A Multimodal Stochastic Planning Approach for Navigation and Multi-Robot Coordination
Mark Gonzales, Ethan Oh, Joseph Moore
AI summary
Problem
Standard sampling-based planners concentrate trajectories around a single best policy, causing robots to get trapped in local minima or deadlock in complex, dynamic environments. Centralized multi-robot coordination methods scale poorly, while distributed approaches lack the trajectory diversity needed for reliable collision avoidance.
Approach
The authors modify the cross-entropy method to optimize a Gaussian Mixture Model policy, using feasibility sampling and K-means clustering to preserve diverse trajectory modes. Robots share these multimodal policies and run a lightweight coordination step to select globally feasible combinations.
Key results
- Preserves multiple trajectory modes to escape local minima
- Enables distributed multi-robot coordination via policy sharing
- Increases success rates in trap environments and collision avoidance
- Validates real-time feasibility through hardware experiments
Why it matters
Provides a scalable, robust planning framework for autonomous robots navigating complex environments and coordinating in teams without centralized computational bottlenecks.
Abstract
In this paper, we present a receding-horizon, sampling-based planner capable of reasoning over multimodal policy distributions. By using the cross-entropy method to optimize a multimodal policy under a common cost function, our approach increases robustness against local minima and promotes effective exploration of the solution space. We show that our approach naturally extends to multi-robot collision- free planning, enables agents to share diverse candidate policies to avoid deadlocks, and allows teams to minimize a global objective without incurring the computational complexity of centralized optimization. Numerical simulations demonstrate that employing multiple modes significantly improves success rates in trap environments and in multi-robot collision avoid- ance. Hardware experiments further validate the approach’s real-time feasibility and practical performance.