// the find
atb033/multi_agent_path_planning
Python implementation of a bunch of multi-robot path-planning algorithms.
A collection of textbook-style Python implementations of multi-agent path planning algorithms: SIPP and CBS for centralized planning, velocity obstacles and NMPC for decentralized/reactive planning, plus a TPG-based post-processing step to turn a discrete plan into an executable schedule. Useful for students or researchers who want working code to go with the papers rather than pseudocode.
Each algorithm is paired with a direct link to its source paper and the code structure maps cleanly onto the paper's concepts, so it reads well alongside the reference. It covers both the centralized (global optimality, higher compute) and decentralized (reactive, local) ends of the MAPF spectrum in one place, which is rare for a single repo. The visualization scripts and GIFs make it easy to sanity-check a planner's output without writing your own plotting code.
No commits since April 2023, so this is effectively frozen and any open issues or incompatibilities with newer numpy/scipy versions won't get fixed. There's no test suite and no packaging (no setup.py or pyproject.toml), so it's a pile of scripts to clone and run from the right subdirectory, not a library you pip install. The repo ships thousands of benchmark YAML files committed directly into the tree instead of fetched on demand, which bloats clone size for no good reason. The CBS and SIPP implementations are reference quality — don't expect symmetry breaking, bypass heuristics, or any of the optimizations that make MAPF solvers viable past toy grid sizes.