Rectify, Don't Regret: Avoiding Pitfalls of Differentiable Simulation in Trajectory Prediction

2026-03-24

AI Summary

This work addresses the limitations of both open-loop trajectory prediction and fully differentiable closed-loop simulation in autonomous driving. Open-loop approaches suffer from error accumulation due to initial deviations, while fully differentiable closed-loop methods often encourage shortcut learning by allowing models to non-causally exploit future information—effectively “regretting” past predictions rather than learning genuine reactive recovery behaviors. To mitigate this, the authors propose a disentangled rolling horizon unrolling mechanism that explicitly severs the computational graph between simulation steps, thereby preventing future information leakage and enforcing causal correction strategies from perturbed states. Evaluated on nuScenes and DeepScenario, the method significantly improves safety: under high replanning frequencies, it reduces target collisions by 33.24% compared to fully differentiable closed-loop baselines and decreases collisions by 27.74% over open-loop methods in dense scenarios, while also enhancing multimodal diversity and lane alignment.

Technology Category

• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Current open-loop trajectory models struggle in real-world autonomous driving because minor initial deviations often cascade into compounding errors, pushing the agent into out-of-distribution states. While fully differentiable closed-loop simulators attempt to address this, they suffer from shortcut learning: the loss gradients flow backward through induced state inputs, inadvertently leaking future ground truth information directly into the model's own previous predictions. The model exploits these signals to artificially avoid drift, non-causally "regretting" past mistakes rather than learning genuinely reactive recovery. To address this, we introduce a detached receding horizon rollout. By explicitly severing the computation graph between simulation steps, the model learns genuine recovery behaviors from drifted states, forcing it to "rectify" mistakes rather than non-causally optimizing past predictions. Extensive evaluations on the nuScenes and DeepScenario datasets show our approach yields more robust recovery strategies, reducing target collisions by up to 33.24% compared to fully differentiable closed-loop training at high replanning frequencies. Furthermore, compared to standard open-loop baselines, our non-differentiable framework decreases collisions by up to 27.74% in dense environments while simultaneously improving multi-modal prediction diversity and lane alignment.

Problem

• trajectory prediction
• differentiable simulation
• shortcut learning
• causality
• error compounding

Innovation

• differentiable simulation
• trajectory prediction
• receding horizon
• causal learning
• autonomous driving