Multi Quadruped Planning

Motivation:

There is work being done on training local locomotion ML policies for complex robots which require non-trivial controllers. These local locomotion policies avoid obstacles while also handling uneven terrain to reach a local way-point. This work focuses on applying these local policies for multiple robots sharing the same workspace. The main objective is to apply Multi-agent path finding algorithms (MAPF) along with pre-trained local policies to plan collision free paths for quadrupeds working collaboratively.

Problem Statement:

The problem boils down to one question - How to leverage these pre-trained policies to reduce constraints in a multi-agent path finding algorithm.