Foresight in Motion: Reinforcing Trajectory Prediction with Reward Heuristics

2025-07-16

AI Summary

To address insufficient accuracy and interpretability in traffic participant trajectory prediction for autonomous driving, this work proposes a “reasoning-before-prediction” paradigm. First, it formulates an intent-aware inverse reinforcement learning (IRL) framework centered on query-based representation, modeling behavioral intent as an interpretable reward distribution. Subsequently, reward-informed policy rollout generates multimodal intent priors to guide trajectory forecasting. The method integrates vectorized scene encoding, query-centered contextual aggregation, and a hierarchical DETR decoder augmented with bidirectional selective state space models (SSMs). Experiments on Argoverse 2 and nuScenes demonstrate significant improvements in prediction confidence and physical plausibility, achieving state-of-the-art (SOTA) accuracy while enhancing model interpretability—offering a novel, safety-critical trajectory prediction framework.

Technology Category

• Planning, Routing, and Scheduling: Model-Based Reasoning
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Motion forecasting for on-road traffic agents presents both a significant challenge and a critical necessity for ensuring safety in autonomous driving systems. In contrast to most existing data-driven approaches that directly predict future trajectories, we rethink this task from a planning perspective, advocating a "First Reasoning, Then Forecasting" strategy that explicitly incorporates behavior intentions as spatial guidance for trajectory prediction. To achieve this, we introduce an interpretable, reward-driven intention reasoner grounded in a novel query-centric Inverse Reinforcement Learning (IRL) scheme. Our method first encodes traffic agents and scene elements into a unified vectorized representation, then aggregates contextual features through a query-centric paradigm. This enables the derivation of a reward distribution, a compact yet informative representation of the target agent's behavior within the given scene context via IRL. Guided by this reward heuristic, we perform policy rollouts to reason about multiple plausible intentions, providing valuable priors for subsequent trajectory generation. Finally, we develop a hierarchical DETR-like decoder integrated with bidirectional selective state space models to produce accurate future trajectories along with their associated probabilities. Extensive experiments on the large-scale Argoverse and nuScenes motion forecasting datasets demonstrate that our approach significantly enhances trajectory prediction confidence, achieving highly competitive performance relative to state-of-the-art methods.

Problem

• Improving trajectory prediction for autonomous driving safety
• Incorporating behavior intentions for spatial guidance
• Enhancing prediction confidence with reward-driven intention reasoning

Innovation

• Query-centric Inverse Reinforcement Learning for intention reasoning
• Hierarchical DETR-like decoder with bidirectional models
• Reward-driven policy rollouts for trajectory generation