Dynamic Gap: Safe Gap-based Navigation in Dynamic Environments

TL;DR

This work extends gap-based local planning to unknown dynamic environments by tracking the evolution of free space in an egocentric frame and by propagating gap dynamics over a short horizon. It integrates gap detection, association, estimation, and propagation with parallel-navigation guidance and Artificial Harmonic Potential Fields to synthesize collision-free trajectories through reachable gaps, plus MPC-based tracking and a projection-based safety filter for non-ideal conditions. A formal proof of collision-free passage is provided under ideal conditions, supported by extensive experiments in both ideal and realistic dynamic scenarios, where Dynamic Gap achieves favorable success rates and outperforms dynamic extensions of competing planners. The approach advances safe navigation in dense and evolving spaces and is open-sourced within Arena-Rosnav for benchmarking and further development.

Abstract

This paper extends the family of gap-based local planners to unknown dynamic environments through generating provable collision-free properties for hierarchical navigation systems. Existing perception-informed local planners that operate in dynamic environments rely on emergent or empirical robustness for collision avoidance as opposed to performing formal analysis of dynamic obstacles. In addition to this, the obstacle tracking that is performed in these existent planners is often achieved with respect to a global inertial frame, subjecting such tracking estimates to transformation errors from odometry drift. The proposed local planner, dynamic gap, shifts the tracking paradigm to modeling how the free space, represented as gaps, evolves over time. Gap crossing and closing conditions are developed to aid in determining the feasibility of passage through gaps, and a breadth of simulation benchmarking is performed against other navigation planners in the literature where the proposed dynamic gap planner achieves the highest success rate out of all planners tested in all environments.

Figures (6)

• Figure 1: Visualization of gaps and trajectories generated by dynamic gap. The central blue circle depicts the ego-robot while the red circle depict dynamic agents. The bold colored arcs labeled A-F are the instantaneous set of detected gaps and the transparent arcs show the predicted gaps obtained by propagating the gap dynamics models, shown as arrows, forward in time. Dashed lines are the candidate trajectories synthesized towards the gap goals shown in yellow. Gap F is predicted to close before the ego-robot can pass through, so it is deemed infeasible and not used during trajectory synthesis.
• Figure 2: Overall workflow for the proposed navigation framework. Red blocks correspond to perceptual modules run at the rate of the laser scanner. Blue blocks are the core planning loop, and green blocks are the lower-level trajectory tracking routine. Dashed outlines represent core contributions.
• Figure 3: (a) Crossing condition for a gap, where between timesteps $t-1$ and $t$, the gap's angular span reaches $0$ rad. (b) Overlapping condition for a gap, where the gap's angular span reaches $2\pi$ rad.
• Figure 4: Diagram for guidance law notation.
• Figure 5: Visualization of four Monte Carlo variations of gaps from Experiment 1. Orange points and lines represent the left side of the gap while red points and lines represent the right side. Solid points and lines represent the original gap geometry whereas hollow points and dashed lines correspond to the inflated version of the gap. Transparent points represent positions at prior timesteps. The blue circle represents the robot along with its finite radius.