I present an adaptive regenerative braking control method for electric cars that combines navigation speed limits, preceding and multiple dynamic targets, road grade, road-surface state, and battery recoverable power. The central motivation is straightforward: an electric car should not rely on a small number of fixed regeneration levels when the road, traffic, battery state, and driver demand are continuously changing. In my method, regenerative braking is treated as a cross-domain control problem rather than a single torque map. The electric car perceives relevant information, decides a safe and comfortable target deceleration, and executes the request through the motor and, only when necessary, through hydraulic compensation. I structure the method around a perception-decision-execution architecture and validate it with road tests under multi-speed-limit and mildly congested conditions. The results show improved energy consumption, fewer brake events, and tighter deceleration consistency while preserving safety redundancy for the electric car.
1. Motivation and Design Position
Fixed regeneration levels are simple and robust, but they create a persistent mismatch between driver expectation and road context in an electric car. When the electric car enters a lower speed-limit zone, follows a slower vehicle, descends a grade, or encounters a wet surface, a fixed drag torque can be either too weak for energy recovery or too strong for comfort and stability. The problem becomes more severe when the battery cannot accept the requested recovery power. In that case, the electric car may lose part of its expected deceleration unless another actuator supplies the missing torque. My design position is therefore that regeneration should be adaptive, but the adaptation must remain bounded, diagnosable, and fail-safe. The electric car should recover as much energy as possible only inside a safety envelope that is defined by target distance, battery limits, road adhesion, and actuator availability.
| Challenge in an electric car | Consequence of fixed regeneration | Control response in my method |
|---|---|---|
| Speed-limit transitions | Late or unnecessary braking; avoidable energy loss | Navigation-based predictive torque reduction before the limit point |
| Car-following in congestion | Frequent brake and accelerator pedal cycles | Relative-distance and relative-speed based following torque |
| Road grade | Weak recovery downhill; excessive drag uphill | Grade-compensated regeneration torque |
| Battery power limitation | Deceleration gap and torque discontinuity | Electro-hydraulic compensation only when recovery capability is insufficient |
| Wet, icy, or uneven road | Wheel slip, jerk, or unstable yaw response | Road-state-specific torque limits and torque-rate limits |
| Signal faults | False scene recognition and unsafe torque requests | Validity judgment, cached transition, and smooth degradation |
I use the term electric car repeatedly because the method depends on properties that are particularly important for an electric car: fast motor torque response, bidirectional power flow, battery charge-power limits, and blended friction braking. The electric car can recover energy only when the battery and motor accept it, and the electric car can maintain deceleration consistency only when the friction brake can fill the gap. My method therefore treats energy recovery and deceleration consistency as two coupled objectives rather than as separate calibrations.
2. Cross-Domain System Architecture
I designed the electric car control system as a distributed perception-decision-execution architecture. The perception layer collects vehicle, road, and traffic information. The decision layer fuses the information, identifies the scene, and calculates the target regenerative torque. The execution layer coordinates motor regeneration, hydraulic compensation, and status feedback. The electric car remains under closed-loop control because the execution layer reports actual torque and availability back to the decision layer. The architecture is intentionally modular so that an electric car can use available sensors and controllers without requiring a single centralized computer.
| Layer | Main function for the electric car | Typical information or actuator |
|---|---|---|
| Perception | Collect vehicle state, road context, traffic targets, and battery boundaries | Navigation, front camera, inertial measurement, battery management, intelligent driving controller |
| Decision | Fuse multi-source signals, identify scene, calculate target deceleration and torque | Vehicle control unit, battery management, intelligent driving controller |
| Execution | Execute motor recovery and hydraulic compensation; provide feedback and warning | Motor controller, integrated electronic brake, instrument cluster |
The electric car uses multiple data sources with different update rates and failure modes. I therefore separate the data into navigation, intelligent-driving target, battery, road-surface, and grade groups. The electric car does not treat every received value as valid. Instead, the decision layer checks validity before using the value in a torque request. This is important because a single invalid speed limit or a single invalid target distance can create an unnecessary deceleration request in an electric car.
| Data source | Information used by the electric car | Representative accuracy or condition |
|---|---|---|
| Intelligent-driving target data | Relative distance, relative speed, multiple dynamic target states | Distance error about +/-0.5 m; relative speed error about +/-0.1 m/s |
| Navigation data | Speed limit, road type, distance to speed-limit point | Limit recognition above 95%; position error below 10 m |
| Battery data | State of charge, allowed charge power, allowed discharge power, recovery capability | Used to bound regenerative torque in the electric car |
| Road-surface data | Speed bump, rough road, wet road, snow or ice | Valid when feature match is at least 80% over about 200 m |
| Grade data | Road slope angle and acceleration state | Used for grade compensation in the electric car |
I define the main controller interaction around a vehicle control unit that calculates recovery torque, enables or disables the adaptive function, and issues hydraulic compensation requests. The electric car uses periodic communication so that torque requests, battery limits, and brake availability remain synchronized. The torque control accuracy target is within +/-5 N m, and the response time target is below 100 ms. The electric car also receives actual hydraulic brake torque so that the decision layer can distinguish between a requested compensation and an achieved compensation.
| Interaction pair | Signal sent by the vehicle control unit | Signal received by the vehicle control unit | Timing target |
|---|---|---|---|
| Vehicle control unit and intelligent-driving controller | Adaptive recovery enable state, target deceleration | Relative distance, relative speed, driving state, multiple-target intrusion information | Period no more than 20 ms |
| Vehicle control unit and navigation system | Adaptive recovery state, display request | Speed limit, road type, navigation validity | Period about 100 ms |
| Vehicle control unit and integrated electronic brake | Hydraulic compensation request, target compensation torque | Hydraulic compensation availability, actual hydraulic brake torque | Response no more than 100 ms |
| Vehicle control unit and battery management system | Regenerative torque request | Allowed charge and discharge power, state of charge, recovery capability | Period about 10 ms |
| Vehicle control unit and instrument cluster | Adaptive recovery state, fault prompt | Display status | Period about 100 ms |
For the electric car, the execution layer has two distinct torque paths. The motor controller converts kinetic energy into electric energy when the battery accepts charge. The integrated electronic brake supplies friction torque when the electric car cannot achieve the requested deceleration through regeneration alone. I keep the hydraulic path as a redundancy path, not as the primary energy-recovery path. In this way, the electric car preserves energy efficiency during normal operation and preserves deceleration consistency during battery-limited operation.
3. Multi-Source Signal Fusion and Validity Judgment
I use a validity-first fusion process. Before the electric car calculates a target torque, each critical signal is checked against communication status, plausibility, and persistence. The electric car then applies standardized parameters such as speed limit, road type, and grade angle. If a signal is invalid, the electric car either holds the last valid value for a bounded time or degrades to a safe default. This prevents a transient communication fault from becoming a false braking request in an electric car.
| Signal group | Validity rule in the electric car | Invalid-handling action |
|---|---|---|
| Preceding-target signal | Allowed state is true and message is valid | If invalid beyond 3 communication cycles, mark untrusted and use cached transition |
| Navigation signal | Allowed state is true and message is valid | If invalid beyond 5 communication cycles, hold the last valid speed limit |
| Road-surface signal | Camera is unobstructed and feature match is at least 80% | Use cached road state and refresh at 500 ms; rescan after timeout |
| Battery signal | Battery message is valid and state-of-charge fluctuation is within +/-1% | Assume recovery capability is limited and enable hydraulic redundancy |
| Grade signal | Sensor or fusion estimate is plausible | Use default grade and reduce adaptation gain in the electric car |
The electric car estimates grade angle from longitudinal acceleration and speed change. I use the following relationship, with positive grade denoting uphill and negative grade denoting downhill:
$$
\theta=\arcsin\left(\frac{a-\dot v}{g}\right)
$$
Here, theta is the road grade angle, a is the measured longitudinal acceleration, v-dot is the speed derivative, and g is gravitational acceleration. For the electric car, grade compensation is not a cosmetic correction. It changes the recovery torque because the same motor torque produces a different deceleration on an uphill road than on a downhill road.
I also normalize the navigation speed limit. If the navigation signal is invalid, the electric car holds the last valid speed limit and compares it with a minimum value of 20 km/h. The initial value is set high to avoid an unintended deceleration request during initialization. The electric car uses road-type encoding for distance estimation and signal validity. I keep the default road type as a county road when the navigation state is not trustworthy, because this default is conservative for distance estimation.
4. Enabling, Exiting, and Degradation Logic
The electric car enables adaptive recovery only when the necessary signals and actuators are available. I require valid preceding-target and navigation signals, a ready vehicle state, an adaptive recovery mode selected by the driver, and either an available hydraulic brake or sufficient battery charge power for the requested torque. The last condition is important because the electric car should not promise a deceleration that neither the motor nor the brake can deliver.
| State | Condition in the electric car | Action |
|---|---|---|
| Enable | Target and navigation signals valid; vehicle ready; adaptive mode selected; brake available or battery power sufficient | Enter adaptive recovery |
| Exit by signal fault | Target or navigation signal becomes untrusted | Exit with highest priority and switch smoothly to medium recovery |
| Exit by driver selection | Driver selects low, medium, or high fixed level | Exit adaptive mode and follow the fixed level |
| Exit by capability limit | Battery charge power insufficient and hydraulic brake unavailable | Exit and degrade to medium recovery |
I define the exit priority so that a signal fault dominates a capability limitation, and a capability limitation dominates a failure to enter the adaptive mode. The electric car uses a smooth transition to avoid a torque step. The transition is especially important in an electric car because the motor can change torque quickly, and an abrupt removal of regeneration can be perceived as a lurch.
$$
p_{\mathrm{exit}}:\;0x2\;\succ\;0x3\;\succ\;0x1
$$
In this expression, 0x2 denotes signal-fault exit, 0x3 denotes recovery-capability-limited exit, and 0x1 denotes inability to enter adaptive recovery. The electric car announces the state change through the instrument cluster and keeps the driver informed without requiring an intrusive warning.
5. Adaptive Regenerative Torque Calculation
I calculate the adaptive recovery request as a multi-objective weighted function. The electric car combines a following-scenario torque, a speed-limit-scenario torque, a grade compensation torque, and a lower torque bound. The lower bound prevents the electric car from requesting excessive deceleration through regeneration alone. The zero upper bound prevents positive traction torque from being generated by the recovery function.
$$
T_{q,\mathrm{Auto}}=\max\left\{T_{q3},\min\left[T_{q1}r_{t1}+T_{q,\mathrm{slope}},\;T_{q2}+T_{q,\mathrm{slope}},\;0\right]\right\}
$$
Here, Tq,Auto is the adaptive recovery torque request, Tq1 is the following-scenario recovery torque, rt1 is the following-torque correction factor, Tq,slope is the grade compensation torque, Tq2 is the speed-limit-scenario recovery torque, and Tq3 is the lower recovery bound. I set Tq3 not lower than -2000 N m, which corresponds to a maximum deceleration not exceeding about 0.2 g. At low speed, the electric car limits the recovery torque to zero to avoid an abrupt stop and to preserve creep behavior.
| Symbol | Meaning in the electric car controller | Typical boundary |
|---|---|---|
| Tq,Auto | Final adaptive recovery torque request | Negative for recovery; zero at low speed |
| Tq1 | Following-scenario torque | Based on relative distance and relative speed |
| Tq2 | Speed-limit-scenario torque | Based on limit value, distance, and current speed |
| Tq,slope | Grade compensation torque | Positive uphill correction, negative downhill correction |
| Tq3 | Lower torque bound | Not lower than -2000 N m |
For the following scenario, I calculate a target deceleration from relative distance and relative speed. The electric car should reduce torque when the distance margin is large and increase recovery when the distance margin is small. The following deceleration is:
$$
a_1=-\frac{\min(V,0)\,V}{2\max\left(L_x-L_s,0.1\right)}
$$
Here, a1 is the following-scenario target deceleration, V is the relative speed, Lx is the relative distance to the preceding target, and Ls is the safe distance. The electric car then converts the target deceleration into a wheel-end torque request:
$$
T_{q1}=\frac{a_1 m r}{i_g i_0 \eta}
$$
Here, m is the vehicle mass, r is the wheel rolling radius, ig is the transmission ratio, i0 is the final-drive ratio, and eta is the drivetrain efficiency. The safe distance is calculated with a reaction-time term and a minimum-distance term:
$$
L_s=\max\left(V_0 t,\;L_0+0.5V_0\right)
$$
Here, V0 is the current speed of the electric car, t is the driver reaction time used by the safety model, and L0 is the minimum safe distance. I keep this distance model conservative because the electric car may use regeneration as a comfort-oriented deceleration source, and the driver should not perceive the system as following too closely.
For the speed-limit scenario, I calculate a target deceleration from the squared speed difference and the available distance to the limit point:
$$
a_2=\frac{\min\left(V_2^2-V_0^2,0\right)}{2\max\left(L_2,L_1\right)}
$$
Here, a2 is the speed-limit-scenario target deceleration, V2 is the speed-limit value including an overspeed offset, V0 is the current speed, L2 is the distance to the speed-limit point, and L1 is the minimum deceleration distance, which I set to 50 m. The corresponding torque request is:
$$
T_{q2}=\frac{a_2 m r}{i_g i_0 \eta}
$$
The electric car uses an overspeed offset to avoid frequent small decelerations caused by speed-limit quantization and navigation error. I define the offset speed as:
$$
V_2=\frac{V_3}{3.6}+V_{\mathrm{offset}}
$$
Here, V3 is the navigation speed limit in km/h and V_offset is the overspeed offset. The electric car estimates the distance to the limit point according to road type. I use a longer preview distance on a highway and a shorter preview distance on a county road. The electric car gives priority to the navigation camera distance when the vehicle is inside an electronic enforcement zone.
| Road type in the electric car | Representative preview distance | Control intention |
|---|---|---|
| Highway | About 600 m | Early and gentle speed-limit adaptation |
| Arterial road | About 300 m | Balanced preview and traffic response |
| County road or other road | About 100 m | Short-range response with conservative torque |
For grade compensation, I adjust the recovery torque according to the slope angle. On an uphill road, I reduce the recovery intensity to lower the drag feeling. On a downhill road, I increase the recovery intensity to capture more energy. The grade compensation torque is:
$$
T_{q,\mathrm{slope}}=\frac{k\theta m g r}{i_g i_0 \eta}
$$
Here, k is the grade compensation coefficient. I use 0.8 for uphill and -1.2 for downhill. The sign convention keeps the electric car consistent: a downhill road supports additional recovery, while an uphill road reduces recovery so that the electric car does not feel overly braked.
The electric car also enforces actuator and stability constraints. The requested torque must satisfy battery charge power, motor capability, and vehicle stability limits. I express the general constraint as:
$$
\left|T_q\right|\le\min\left(T_{\mathrm{motor}},T_{\mathrm{batt}},T_{\mathrm{stability}}\right)
$$
For battery-limited operation, the available recovery torque can be approximated from the allowed charge power and motor speed:
$$
T_{\mathrm{batt}}=\frac{P_{\mathrm{chg,allow}}}{\omega_m\eta_g}
$$
Here, Pchg,allow is the allowed battery charge power, omega_m is the motor speed, and eta_g is the generator-path efficiency. The electric car uses this value as a dynamic upper bound on recovery. If the following-scenario demand conflicts with the speed-limit demand, I prioritize the following-scenario torque to preserve longitudinal safety. The electric car should not insist on a navigation speed limit when a closer target requires a stronger response.
$$
T_q=T_{q1},\quad \text{if the following demand is more urgent than the limit demand}
$$
6. Electro-Hydraulic Coordination
I treat hydraulic compensation as a safety redundancy for the electric car. The hydraulic brake does not participate in normal energy recovery unless the battery cannot accept the requested recovery torque. This distinction is essential because unnecessary hydraulic blending reduces energy efficiency, while missing hydraulic compensation reduces deceleration consistency. The electric car therefore compares the target recovery torque with the available recovery torque and activates hydraulic compensation only when the difference exceeds a calibrated threshold.
$$
T_{q,\mathrm{diff}}=\left|T_{q,\mathrm{Auto}}-T_{q,\min}\right|
$$
Here, Tq,diff is the torque difference between the target recovery torque and the available recovery torque. The electric car allocates the compensation torque mainly to the front axle and the remainder to the rear axle:
$$
T_{q,\mathrm{front}}=0.7T_{q,\mathrm{diff}}
$$
$$
T_{q,\mathrm{rear}}=0.3T_{q,\mathrm{diff}}
$$
The total braking torque is the sum of motor recovery torque and hydraulic compensation torque:
$$
T_{q,\mathrm{total}}=T_{q,\mathrm{MGU}}+T_{q,\mathrm{hyd}}
$$
I also limit the rate of hydraulic torque change so that the electric car does not experience a sudden deceleration step:
$$
\left|\frac{dT_{q,\mathrm{hyd}}}{dt}\right|\le R_{\mathrm{hyd}}
$$
Here, R_hyd is the calibrated hydraulic torque rate limit. The exact value depends on the brake system and the vehicle state. The electric car uses a lower rate on a low-adhesion surface and a higher rate on a dry surface. The compensation logic is shown in the following table.
| Compensation condition | Requirement in the electric car | Resulting action |
|---|---|---|
| Recovery capability sufficient | Available recovery torque meets the target | Motor recovery only |
| Recovery capability insufficient | Target recovery exceeds available recovery by a threshold | Motor recovery plus hydraulic compensation |
| Hydraulic brake unavailable | Brake fault or unavailable state | Limit recovery request and degrade the electric car to a safe mode |
| Special gear transition | Drive or reverse gear change in progress | Inhibit compensation until the transition is complete |
| Adaptive recovery disabled | Driver selects a fixed level or system exits adaptive mode | No adaptive compensation request |
I designed the compensation trigger to require four conditions at the same time: the absolute target recovery torque exceeds the absolute available recovery torque, the hydraulic brake is available without a fault, the electric car is not in a special gear-change state, and the adaptive recovery function is enabled. These conditions prevent hydraulic compensation from becoming a hidden default path. The electric car should use hydraulic torque to fill a deceleration gap, not to mask a persistent recovery limitation.
7. Multi-Scenario Control Strategies
I use scene-specific strategies because a single torque formula cannot cover all electric car operating conditions. The electric car identifies following, speed-limit, grade, abnormal-road, multi-target, and special-state scenarios. Each scene modifies the target torque or the torque-rate limit. The common principle is that the electric car should recover energy continuously, but it should never trade stability for recovery.
7.1 Abnormal Road Control
For an uneven road, the electric car reduces the target speed and limits recovery torque to avoid a sudden drag over a speed bump or rough patch. For a wet road, the electric car limits recovery capability and torque rate to reduce wheel-slip risk. For a snow or ice road, the electric car switches to a snow-like recovery calibration and uses a lower torque rate. These strategies are especially important for an electric car because motor torque can be applied faster than a driver expects.
| Road state | Electric car recovery action | Torque or speed limit |
|---|---|---|
| Uneven road or speed bump | Reduce target speed; limit recovery; inhibit hydraulic compensation for non-decoupled pedal architectures | Target speed below 20 km/h; recovery not below -500 N m |
| Wet road | Limit battery recovery capability and smooth torque changes | Recovery within 60% of maximum; torque rate no more than 50 N m/s |
| Snow or ice road | Use snow-like recovery calibration | Torque rate no more than 30 N m/s |
| Normal dry road | Use full adaptive recovery envelope | Subject to motor, battery, and stability limits |
7.2 Multiple Dynamic Targets
I use multiple dynamic target information to distinguish a normal preceding vehicle from a target that may intrude into the electric car path. The electric car sorts targets by distance and evaluates crossing risk. For a crossing target, I compare the target crossing time with the ego arrival time. If the ego arrival time is less than or equal to the target crossing time plus a one-second margin, the electric car classifies the target as dangerous and prioritizes regenerative deceleration. If the battery cannot supply the required recovery torque, the electric car requests hydraulic compensation.
$$
t_1=\frac{d_{\mathrm{target}}}{v_{\mathrm{target}}}
$$
$$
t_2=\frac{d_{\mathrm{ego}}}{v_{\mathrm{ego}}}
$$
$$
\text{Risk condition: }t_2\le t_1+t_0,\quad t_0=1\;\mathrm{s}
$$
For a target traveling along the same lane, the electric car treats it as a preceding vehicle and adjusts recovery according to relative distance and relative speed. For a non-intruding target, the electric car does not issue an unnecessary deceleration request. This differentiation reduces false interventions and improves the driving experience of the electric car.
| Target category | Recognition in the electric car | Control response |
|---|---|---|
| Preceding vehicle | Same lane, stable relative motion | Following-scenario recovery torque |
| Dangerous crossing target | Arrival time condition satisfied | Priority regenerative deceleration; hydraulic compensation if needed |
| Safe crossing target | Arrival time condition not satisfied | No extra recovery request; continue normal control |
| Adjacent-lane target | Possible intrusion but not yet in path | Monitor and prepare a bounded recovery request |
7.3 Special Operating Conditions
I include special conditions because the electric car may have full battery, low temperature, or gear-transition states in which recovery capability is reduced. In a full-battery or low-temperature state, the electric car precharges hydraulic brake readiness to 1.2 times the demanded torque so that deceleration consistency is maintained when recovery is unavailable. In a drive-to-reverse or reverse-to-drive transition, the electric car inhibits compensation and recovery requests until the transition is complete.
| Special condition | Electric car behavior | Safety objective |
|---|---|---|
| Full battery or low temperature | Prepare hydraulic brake torque; compensate recovery shortage | Maintain deceleration consistency |
| Drive or reverse gear transition | Inhibit recovery and hydraulic compensation requests | Avoid unintended torque direction |
| Low-speed operation | Set recovery torque to zero below the low-speed threshold | Preserve creep and avoid abrupt stop |
| Driver braking override | Yield to driver pedal demand | Keep driver authority in the electric car |
| Signal degradation | Switch to medium recovery and notify the driver | Provide predictable behavior |
I also include a human-machine coordination prompt. When the electric car is in drive, the speed is at least 10 km/h, the accelerator pedal opening is at least 5%, and the adaptive torque request is less than or equal to -50 N m, the instrument cluster suggests that the driver reduce speed. The prompt lasts two seconds and repeats no earlier than ten seconds later. This prompt is not a substitute for control; it helps the driver understand why the electric car is preparing to recover energy.
$$
\text{Prompt}=\left(\mathrm{gear}=D\right)\land\left(v\ge10\;\mathrm{km/h}\right)\land\left(\alpha_{\mathrm{pedal}}\ge5\%\right)\land\left(T_{q,\mathrm{Auto}}\le-50\;\mathrm{N\,m}\right)
$$
8. Road Validation
I validated the method with road tests using two electric cars of the same configuration. One electric car ran with the adaptive recovery function enabled, and the other ran with the function disabled. I kept the battery state, tire pressure, climate settings, and route identical as far as practicable. The route included highway, urban elevated road, and urban road sections. It covered following, speed-limit, grade, and mildly congested conditions. I ran three test rounds and recorded state-of-charge consumption, brake events, accelerator events, deceleration fluctuation, and yaw response.
| Test item | Condition in the electric car test |
|---|---|
| Vehicle configuration | Two electric cars of the same type; one adaptive, one baseline |
| Vehicle mass | About 2540 kg |
| Battery capacity | About 25.5 kWh |
| Motor maximum power | About 160 kW |
| Maximum recovery torque | About 2000 N m |
| Initial state of charge | 80% for both electric cars |
| Route length | About 32 km, including highway, elevated road, and urban road |
| Test rounds | Three rounds with 30-minute intervals |
| Environment | Cloudy, 13 to 15 degrees Celsius, dry road except special tests |
I calculate the energy improvement rate from the state-of-charge consumption of the baseline electric car and the adaptive electric car:
$$
\eta_E=\frac{\Delta SOC_{\mathrm{off}}-\Delta SOC_{\mathrm{on}}}{\Delta SOC_{\mathrm{off}}}\times100\%
$$
I also calculate the brake-event reduction and accelerator-event reduction as follows:
$$
R_b=\frac{N_{b,\mathrm{off}}-N_{b,\mathrm{on}}}{N_{b,\mathrm{off}}}\times100\%
$$
$$
R_a=\frac{N_{a,\mathrm{off}}-N_{a,\mathrm{on}}}{N_{a,\mathrm{off}}}\times100\%
$$
$$
R_t=\frac{T_{b,\mathrm{off}}-T_{b,\mathrm{on}}}{T_{b,\mathrm{off}}}\times100\%
$$
Here, N_b is the number of brake events, N_a is the number of accelerator events, and T_b is the total brake-pedal duration. The electric car with adaptive recovery showed lower state-of-charge consumption in every round, although the improvement depended on the route mix. The third round contained more urban congestion and following events, so the predictive deceleration and dynamic recovery strategy had more opportunity to reduce unnecessary brake-accelerate cycles.
| Test round | State-of-charge consumption with adaptive recovery | State-of-charge consumption without adaptive recovery | Energy improvement |
|---|---|---|---|
| 1 | 15.50% | 15.70% | 1.27% |
| 2 | 15.15% | 15.80% | 4.11% |
| 3 | 15.20% | 16.25% | 6.46% |
| Average | 15.28% | 15.92% | 4.02% |
The average improvement was 4.02%. I interpret this result carefully. The electric car does not recover more energy simply because the torque request is larger. The improvement comes from better timing: the electric car begins recovery earlier when a speed limit or a slower target is predicted, and it avoids some friction-brake events that would otherwise dissipate kinetic energy. The electric car also reduces unnecessary acceleration after a late brake event, which further lowers consumption.
| Metric | Adaptive electric car | Baseline electric car | Change |
|---|---|---|---|
| Driving duration | 3249.8 s | 3293.7 s | Shorter with adaptive recovery |
| Brake events | 62 | 112 | 44.64% fewer |
| Accelerator events | 136 | 436 | 68.81% fewer |
| Total brake-pedal duration | 252.6 s | 500.5 s | 49.53% shorter |
| Accelerator-pedal opening variance | 186.9 | 230.9 | Lower pedal activity |
I use deceleration fluctuation as a consistency metric. The standard deviation of longitudinal acceleration is:
$$
\sigma_a=\sqrt{\frac{1}{N}\sum_{i=1}^{N}\left(a_i-\bar a\right)^2}
$$
Here, a_i is the sampled longitudinal acceleration, a-bar is the mean acceleration, and N is the number of samples. The adaptive electric car kept deceleration fluctuation within about +/-0.02 g. The acceleration variation range was reduced by about 33.8% relative to the baseline. The wheel-end torque variation was reduced by about 9.87%. These results support the claim that the electric car can recover energy without sacrificing deceleration continuity.
| Performance item | Observed result for the electric car | Interpretation |
|---|---|---|
| Average energy improvement | 4.02% | Better recovery timing and fewer loss cycles |
| Brake-event reduction | 44.64% | Predictive deceleration reduces friction braking |
| Accelerator-event reduction | 68.81% | Less brake-accelerate oscillation |
| Brake-pedal duration reduction | 49.53% | More recovery is used before friction braking |
| Deceleration fluctuation | Within +/-0.02 g | Consistent deceleration in the electric car |
| Acceleration variation reduction | About 33.8% | Smoother longitudinal response |
| Torque variation reduction | About 9.87% | More continuous recovery torque |
I also validated special conditions in which the electric car had no recoverable battery power. In this case, the hydraulic path supplied the deceleration gap while the motor path remained at its available limit. The electric car did not show an obvious recovery drag step or low-speed rollback. The yaw response on a low-adhesion turn remained stable. This is the intended behavior of the redundancy design: the electric car uses hydraulic compensation only when the battery cannot accept recovery, and the compensation is blended so that the total deceleration remains continuous.
9. Engineering Discussion
I consider the method practical for an electric car because it separates perception, decision, and execution. The electric car can reuse existing navigation, camera, battery, and brake signals. The main calibration effort is in validity thresholds, torque-rate limits, and scene-priority rules. I do not require a full high-definition map for every road. The electric car can operate with navigation speed limits and camera-based road state, and it can degrade gracefully when those signals are unavailable.
| Calibration area | Why it matters for the electric car | Representative tuning direction |
|---|---|---|
| Signal timeout | Prevents false scene changes in the electric car | Short timeout for target data; longer timeout for navigation |
| Safe-distance model | Defines following recovery aggressiveness | Conservative at high speed; responsive at low speed |
| Speed-limit preview | Controls early recovery before a limit point | Longer preview on highway; shorter on urban road |
| Grade coefficient | Balances drag feeling and energy recovery | Reduce recovery uphill; increase recovery downhill |
| Hydraulic torque rate | Maintains comfort during compensation | Lower rate on wet or icy road |
| Low-speed cutoff | Preserves creep and avoids abrupt stop | Set recovery to zero below a calibrated speed |
I also examined failure behavior. The electric car should not interpret a missing navigation message as a speed-limit drop to zero. The electric car should not interpret a missing target message as a clear road. The electric car should not interpret a battery communication fault as unlimited charge power. My validity logic handles these cases by holding, defaulting, or degrading. The electric car also gives priority to the driver pedal demand. If the driver presses the brake or accelerator, the driver command overrides the adaptive recovery request.
| Failure mode in the electric car | Possible consequence without handling | Mitigation in my method |
|---|---|---|
| Navigation timeout | False speed-limit deceleration | Hold last valid limit and require validity before use |
| Target timeout | Missed following deceleration | Use cached transition and degrade to medium recovery |
| Battery power fault | Recovery torque step or deceleration gap | Assume limited recovery and enable hydraulic redundancy |
| Brake unavailable | Inability to fill deceleration gap | Limit adaptive recovery and prompt the driver |
| Road-state misclassification | Excessive torque on low-adhesion road | Use feature-match threshold and torque-rate limits |
I believe the method is scalable to different electric car architectures. A rear-drive electric car can use the same decision layer with a different hydraulic split. An all-wheel-drive electric car can distribute recovery torque between axles while keeping the same validity and compensation framework. A plug-in hybrid electric car can use the same scene logic, although the engine-start and battery-boundary states require additional calibration. The common element is the electric car’s need to coordinate recovery with friction braking and to preserve deceleration consistency when recovery is limited.
10. Conclusion
I proposed and validated an adaptive regenerative braking method for electric cars based on multi-source information fusion. The method uses a perception-decision-execution architecture, a validity-first signal fusion process, a dynamic recovery torque calculation, and an electro-hydraulic compensation strategy that activates only when battery recovery capability is insufficient. The electric car therefore recovers energy when possible and preserves deceleration consistency when recovery is constrained. The road results show a 4.02% average energy improvement, a 44.64% reduction in brake events, and deceleration fluctuation within +/-0.02 g under multi-speed-limit and mildly congested conditions.
I draw three conclusions. First, an electric car benefits more from predictive recovery timing than from a uniformly stronger fixed drag. Second, hydraulic compensation is valuable only when it is treated as a bounded redundancy path; otherwise, it can erase the energy benefit of the electric car. Third, signal validity and degradation logic are not secondary details. They are central to making adaptive recovery trustworthy in an electric car. The electric car should always prefer a predictable safe response over an aggressive but fragile recovery request.
| Design objective | Mechanism in the electric car | Validated outcome |
|---|---|---|
| Improve energy efficiency | Navigation and target preview; grade compensation | 4.02% average energy improvement |
| Reduce brake activity | Predictive deceleration and dynamic recovery torque | 44.64% fewer brake events |
| Maintain deceleration consistency | Motor recovery plus hydraulic gap filling | Fluctuation within +/-0.02 g |
| Preserve safety redundancy | Validity judgment, degradation, and torque-rate limits | Stable response in special conditions |
| Support practical deployment | Distributed architecture and bounded calibration | Compatible with existing electric car controllers |
For future work, I plan to improve the prediction of preceding-vehicle behavior and road state with learning-based models while keeping the safety envelope deterministic. I also plan to adapt hydraulic compensation ratio and response speed to battery aging and motor efficiency. Another direction is to extend the same electric car control framework to higher-level driver assistance, smart charging coordination, and personalized driving modes. The long-term goal is a general electric car control layer that balances energy recovery, deceleration consistency, comfort, and safety across different electrified vehicle architectures.