Organizations: Key Laboratory of Road and Traffic Engineering of the Ministry of Education, Tongji University, Shanghai, China · College of Automotive and Energy Engineering, Tongji University, Shanghai, China
Learning-based vehicle motion control requires coordination between diverse task objectives and state-dependent chassis safety constraints. Conventional cascaded architectures may compromise task performance when safety filters modify nominal commands independently of policy learning. This article proposes a differentiable chassis safety layer that modularly integrates task-oriented control with explicit physical constraints. Actuator limits, individual-tire force constraints, and vehicle handling envelopes are incorporated into sequential quadratic programs. Each projected command propagates the predicted vehicle dynamics and updates subsequent constraints, enabling temporal gradient coupling through implicit differentiation. A two-stage learning strategy combines expert imitation with risk-guided supervision and conditional intervention guidance to coordinate nominal control decisions with chassis safety constraints. Closed-loop evaluations under unseen scenarios, emergency maneuvers, and ablation studies demonstrate that the proposed method improves safety-performance coordination and exploits vehicle dynamic potential while maintaining handling stability.
Figures & tables
Figure 1: Motivation for policy–safety coordination. (a) Handling demand versus physical limits. (b) Nominal-to-executed steering mismatch after safety projection. (c) Coordination of steering policy learning and state-dependent safety correction. The admissible sets are schematic.
Figure 2: Architecture of the proposed safety-guarded task learning framework.
Parameter
Value
Network Architecture
Input dimension
22+4H
Preview/control horizon (H,Hc)
(10,10)
Hidden dimension
192
Residual blocks
3
Dropout rate
0.05
Table 1: Key network and training hyperparameters.
Parameter
Symbol
Value
Unit
Vehicle mass
m
1259.9
kg
Yaw moment of inertia
Iz
1343.1
kg ⋅ m 2
CG to front axle
lf
1.04
m
CG to rear axle
lr
1.56
m
Front cornering stiffness per tire
Cf
39800
N/rad
Rear cornering stiffness per tire
Cr
34050
N/rad
Table 2: Vehicle and steering-actuator parameters.
Metric
LQR
MPC
Ours
Trajectory generalization
Maximum absolute lateral error (m)
0.1482
0.1564
0.0878
Mean absolute lateral error (m)
0.0342
0.0358
0.0199
Maximum absolute sideslip angle (deg)
1.2078
1.1229
2.0003
Mean absolute sideslip angle (deg)
0.1852
0.1675
0.3732
Maximum absolute yaw rate (deg/s)
11.2793
11.6960
13.6922
Table 3: Statistical comparison of tracking accuracy and handling stability in trajectory, speed, and tire–road friction generalization tests.
Figure 3: Safety validation of the proposed controller under near-limit maneuvers: (a) trajectory-tracking result; (b) sideslip angle and yaw rate; (c) front and rear steering commands; (d) evolution of the dynamic stability envelope; (e) stability-boundary exceedance; (f) evolution of the tire-adhesion safety margin.
Metric
Policy Network
Policy Network + Non-diff Safety Layer
Policy Network + Diff Safety Layer + No Joint Training
Ours
Maximum Absolute Lateral Error (m)
0.4562
–
1.8739
0.3422
Mean Absolute Lateral Error (m)
0.1150
–
0.4981
0.1011
Maximum ∣β∣ (deg)
7.9939
–
9.5251
8.5579
Mean ∣β∣ (deg)
1.6558
–
3.5399
2.4303
Maximum ∣γ∣ (deg/s)
23.6297
–
14.0890
25.6503
Mean ∣γ∣ (deg/s)
4.7406
–
4.3293
4.7711
Table 4: Quantitative results of the ablation study under high-speed emergency collision-avoidance maneuvers.
Safety-critical control designs controllers that keep physical systems within hard safety constraints despite uncertainty, disturbances, and limited actuation.
The University of Hong Kong, Pokfulam, Hong Kong, China · Alibaba, Beijing, China · University of Chinese Academy of Sciences, No.1 Yanqihu East Rd, Huairou District, Beijing, China