ECAHD: Efficient Collision-Aware Hierarchical Diffusion Navigation
Jinu Pahk, Theo Taeyeong Kim, Jun Ki Lee, Byoung-Tak Zhang
AI summary
Problem
Existing diffusion-based navigation planners struggle to balance fast global path generation with precise, robot-shape-aware local collision avoidance, often suffering from slow sampling speeds or performance degradation when collision guidance is applied globally.
Approach
ECAHD separates planning into a Sparse Diffusion module for rapid global trajectory generation and a Dense Diffusion module for local refinement, applying a differentiable collision cost only when potential obstacles are detected.
Key results
- Hierarchical diffusion architecture separates global planning from local refinement
- Differentiable collision cost explicitly models robot footprint without extra training
- Over 50% reduction in collision rates with ~1.3% higher success rate in maze2d-large
- Nearly 50% faster inference time and lowest collision rates on D4RL benchmarks
Why it matters
Provides a scalable, real-time navigation framework that enables safe and efficient deployment of learning-based robots in complex, obstacle-dense environments.
Abstract
Safe and fast robot navigation is crucial for deploying robots in real-time interactive environments. Recent diffusion-based approaches for path planning and navigation leverage data-driven learning to generalize across diverse tasks and to quickly generate trajectories, but research explicitly incorporating collision awareness remains limited. In this work, we propose Efficient Collision-Aware Hierarchical Diffusion Navigation (ECAHD), a hierarchical diffusion-based framework designed for both safety and computational efficiency. ECAHD generates a sparse trajectory for global path planning and a dense trajectory for local path refinement. The robot follows a rapidly sampled sparse global trajectory, and when a potential collision is detected, a collision-aware guidance diffusion mech- anism—which accounts for the robot’s shape—adjusts the local trajectory accordingly. Conventional full-sequence diffusion planners suffer from slow sampling speeds and performance degradation when collision-aware guidance is applied across the entire trajectory. ECAHD addresses these issues by significantly reducing the number of waypoints predicted by the global diffusion planner, while delegating robot shape aware collision guidance to the local diffusion planner. This separation not only accelerates planning but also preserves global trajectory quality, as goal-conditioned sampling is no longer disrupted by collision-related constraints. Furthermore, ECAHD allows for increasing the number of global trajectory samples to enhance performance, without incurring substantial computational over- head. In maze2d-large planning tests, ECAHD improved success rates by approximately 1.3% while reducing collision rates by more than 50%, all while cutting inference time by nearly half. In D4RL Navigation benchmark, ECAHD achieved the fastest goal-reaching time and the lowest collision rate among compared methods, demonstrating its effectiveness in efficient and safe robot navigation.