CoPlanner: An Interactive Motion Planner with Contingency-Aware Diffusion for Autonomous Driving
Ruiguo Zhong, Ruoyu Yao, Pei Liu, Xiaolong Chen, Rui Yang, Jun Ma
AI summary
Problem
Current autonomous driving planners either decouple prediction and planning, causing social inconsistency, or collapse to a single most-likely future, leaving vehicles unprepared for uncertainty and lacking fallback strategies.
Approach
The method uses a diffusion-based framework that first generates a validated shared short-term trajectory segment for all agents, then samples diverse long-term branches via pivot-conditioned diffusion, and finally scores ego plans across these branches to balance safety, progress, and comfort.
Key results
- Unified generation-then-evaluation framework for joint multi-agent trajectory and ego planning
- Pivot-conditioned diffusion mechanism enforcing shared short-term consistency with diverse long-term branching
- Contingency-aware multi-scenario scoring strategy mitigating over-conservatism and selecting robust plans
- State-of-the-art closed-loop performance on nuPlan Val14 and Test14 benchmarks with improved safety and comfort metrics
Why it matters
Provides autonomous driving systems with robust, fallback-ready planning capabilities that maintain stability and safety in highly interactive and uncertain traffic environments.
Abstract
Accurate trajectory prediction and motion plan- ning are crucial for autonomous driving systems to navigate safely in complex, interactive environments characterized by multimodal uncertainties. However, current generation-then- evaluation frameworks typically construct multiple plausible trajectory hypotheses but ultimately adopt a single most likely outcome, leading to overconfident decisions and a lack of fallback strategies that are vital for safety in rare but crit- ical scenarios. Moreover, the usual decoupling of prediction and planning modules could result in socially inconsistent or unrealistic joint trajectories, especially in highly interactive traffic. To address these challenges, we propose a contingency- aware diffusion planner (CoPlanner), a unified framework that jointly models multi-agent interactive trajectory generation and contingency-aware motion planning. Specifically, the pivot- conditioned diffusion mechanism anchors trajectory sampling on a validated, shared short-term segment to preserve tem- poral consistency, while stochastically generating diverse long- horizon branches that capture multimodal motion evolutions. In parallel, we design a contingency-aware multi-scenario scoring strategy that evaluates candidate ego trajectories across multi- ple plausible long-horizon evolution scenarios, balancing safety, progress, and comfort. This integrated design preserves feasible fallback options and enhances robustness under uncertainty, leading to more realistic interaction-aware planning. Extensive closed-loop experiments on the nuPlan benchmark demonstrate that CoPlanner consistently surpasses state-of-the-art methods on both Val14 and Test14 datasets, achieving significant im- provements in safety and comfort under both reactive and non- reactive settings. The source code is available on GitHub1.