ROMar 13

FLUX: Accelerating Cross-Embodiment Generative Navigation Policies via Rectified Flow and Static-to-Dynamic Learning

arXiv:2603.1280677.6
AI Analysis

This work addresses the need for comprehensive evaluation and efficient, generalizable navigation policies in robotics, representing a novel method for a known bottleneck rather than an incremental improvement.

The authors tackled the problem of fragmented evaluation in autonomous navigation by introducing DynBench, a dynamic navigation benchmark, and proposed FLUX, a flow-based unified navigation policy that improves per-step inference efficiency by 47% over prior flow-based methods and 29% over diffusion-based ones, achieving state-of-the-art performance across six tasks and zero-shot sim-to-real transfer on multiple platforms.

Autonomous navigation requires a broad spectrum of skills, from static goal-reaching to dynamic social traversal, yet evaluation remains fragmented across disparate protocols. We introduce DynBench, a dynamic navigation benchmark featuring physically valid crowd simulation. Combined with existing static protocols, it supports comprehensive evaluation across six fundamental navigation tasks. Within this framework, we propose FLUX, the first flow-based unified navigation policy. By linearizing probability flow, FLUX replaces iterative denoising with straight-line trajectories, improving per-step inference efficiency by 47% over prior flow-based methods and 29% over diffusion-based ones. Following a static-to-dynamic curriculum, FLUX initially establishes geometric priors and is subsequently refined through reinforcement learning in dynamic social environments. This regime not only strengthens socially-aware navigation but also enhances static task robustness by capturing recovery behaviors through stochastic action distributions. FLUX achieves state-of-the-art performance across all tasks and demonstrates zero-shot sim-to-real transfer on wheeled, quadrupedal, and humanoid platforms without any fine-tuning.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes