Resilient Execution for Mixed Fleets via Spatio-temporal Bounds
2026-02-05T15:00:00Z
When moving a diverse fleet of robots—varying in size and speed—the biggest challenge isn’t just finding a path; it’s surviving the execution. Standard Multi-Agent Pathfinding (MAPF) plans often break the moment a human walks into a corridor or a robot encounters a random obstacle, usually requiring a “Global Stop” or a heavy re-calculation.
In this session, we present a workflow for heterogeneous fleets that decouples high-level coordination from low-level local navigation. We start with a specialized planner (Heterogeneous PIBT) that accounts for different robot footprints to generate an Action Dependency Graph (ADG)—the logical “who goes first” of the fleet.
To make this plan resilient, we “wrap” these dependencies in Spatio-temporal Allocations. Instead of a fixed path, each robot is granted a dynamic “spatial envelope” or “safety bubble.” This allows the individual ROS local planner (like Nav2 or a custom controller) the freedom to divert and maneuver around local obstacles without violating the global plan or causing a deadlock. We will discuss how this method provides a robust framework for interoperability, allowing diverse robots to share space safely even when the environment changes.
These methods will play a major role in the traffic management of Next Generation Open-RMF, allowing AMRs to utilize the full strength of their free space navigation abilities while still respecting the overall flow of traffic.