RAY-TOLD: Ray-Based Latent Dynamics for Dense Dynamic Obstacle Avoidance with TDMPC

2026-04-30

AI Summary

This work addresses the challenge of safe navigation in dense, dynamic crowds, where conventional reactive planners often fall into local optima due to limited prediction horizons. To overcome this limitation, the authors propose a hybrid planning framework that integrates Model Predictive Path Integral (MPPI) control with reinforcement learning. The approach employs a LiDAR-driven latent dynamics model to encode environmental information and combines a learned policy prior with a terminal value function to guide trajectory generation. A novel policy-blending sampling mechanism is introduced to incorporate long-horizon learned intentions into MPPI’s candidate trajectories. Evaluated in high-density stochastic dynamic environments, the method significantly enhances navigation safety and robustness while preserving kinematic feasibility, achieving substantially lower collision rates than the MPPI baseline and demonstrating superior obstacle avoidance performance.

Abstract

Dense, dynamic crowds pose a persistent challenge for autonomous mobile robots. Purely reactive planning methods, such as Model Predictive Path Integral (MPPI) control, often fail to escape local minima in complex scenarios due to their limited prediction horizon. To bridge this gap, we propose Ray-based Task-Oriented Latent Dynamics (RAY-TOLD), a hybrid control architecture that integrates obstacle information into latent dynamics and utilizes the robustness of physics-based MPPI with the long-horizon foresight of reinforcement learning. RAY-TOLD leverages a LiDAR-centric latent dynamics model to encode high-dimensional sensor data into a compact state representation, enabling the learning of a terminal value function and a policy prior. We introduce a policy mixture sampling strategy that augments the MPPI candidate population with trajectories derived from the learned policy, effectively guiding the planner towards the goal while maintaining kinematic feasibility. Extensive tests in a stochastic environment with high-density dynamic obstacles demonstrate that our method outperforms the MPPI baseline, reducing the collision rate. The results confirm that blending short-horizon physics-based rollouts with learned long-horizon intent significantly enhances navigation reliability and safety.

Problem

• dynamic obstacle avoidance
• autonomous mobile robots
• dense crowds
• local minima
• navigation safety

Innovation

• Ray-based latent dynamics
• TDMPC
• MPPI
• policy mixture sampling
• dense dynamic obstacle avoidance