Driving · 2026-10-14

Beyond Planning: Differentiable Control-in-the-loop End-to-End Autonomous Driving

Zhidong Wang, Zirui Li, Zitong Shan, Zhiqi Mao, Qihang Huang, Zhan Chen, Jingsong Liang, Han Yu, Chen Lv

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

DIVA (Differentiable Identified Vehicle dynamics for end-to-end Autonomous driving) embeds an identified actuator-level vehicle dynamics model into an end-to-end autonomous driving training loop, enabling a differentiable chain from perception to vehicle states. It replaces the external controller with a learned control decoder supervised on rolled-out vehicle states, uses rate-consistent control-input knots (3.8 ms control-only cycle), and applies on-policy re-identification plus an online disturbance observer for mismatch robustness.

Why it matters

The authors target the planning-to-control discontinuity in end-to-end autonomous driving, where hand-tuned tracking controllers and non-differentiable vehicle dynamics block gradient flow and deployment suffers from rate mismatch and vehicle-model discrepancies. DIVA connects supervision and gradients to vehicle-state rollouts through a differentiable identified dynamics model, aligning training and deployment under rate-consistent asynchronous execution and mismatch-aware control.

Method

Embed a differentiable actuator-level vehicle dynamics model in the training loop, so the loss/gradients propagate from control decoder outputs back through vehicle states into the network.
Train a learned control decoder supervised on rolled-out vehicle states via dynamics-in-the-loop imitation learning, bypassing aliased control-input label supervision.
Use rate-consistent control-input knots (FOH interpolation) to reconcile label/perception/control rates for asynchronous deployment, plus on-policy re-identification and an online disturbance observer.

Limitation

The method is described for end-to-end autonomous driving in simulation (e.g., Bench2Drive benchmark and CARLA 0.9.15 mentioned in the appendix); future work is stated to transfer the framework to real vehicles.

Abstract (from arXiv)

End-to-end autonomous driving has achieved remarkable progress by planning trajectories directly from raw sensor data. Yet the step from planning to control still relies on hand-tuned tracking controllers that block the gradient path from vehicle motion back to the network. This gap stems from non-differentiable vehicle dynamics, aliased supervision from coarsely sampled expert labels, rate mismatch between high-rate control and low-rate re-planning, and vehicle model discrepancies between training and deployment. This paper proposes DIVA (Differentiable Identified Vehicle dynamics for end-to-end Autonomous driving) to close this gap. An identified actuator-level vehicle model is embedded in the training loop of the end-to-end driving network, forming a differentiable chain from perception to vehicle states. A learning-based control decoder replaces the external controller and is supervised on the rolled-out vehicle states, bypassing the aliased control-input labels. A rate-consistent control-input parameterization reconciles the heterogeneous sampling rates and enables asynchronous execution with a 3.8 ms control-only cycle. On-policy re-identification and an online disturbance observer address training-time distribution shift and deployment-time model mismatch respectively. On the closed-loop Bench2Drive benchmark, DIVA achieves a driving score of 86.7, surpassing the widely adopted state-of-the-art approach by 3.3 points while nearly halving mean jerk and reducing brake switching by 64%, and maintains superior path tracking accuracy under significant vehicle model mismatch at deployment.

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-13 · Indexed by Pub-AI: 2026-10-14 UTC

preprint