Researchers led by Zheyu Tao have built PIBT*T, a planner that finds paths for large fleets of mobile robots while allowing for the time they need to turn, according to a paper on arXiv. Modelling turns makes such planning more realistic but much slower for search-based methods.
The planner builds on priority inheritance with backtracking (PIBT), which plans one time step at a time, and adds ways to handle the dependencies that turns create, local deadlocks, livelock and robots repeatedly pushing each other.
On standard benchmarks it solved problems with up to 1,000 robots within seconds. In a simulated flexible manufacturing system, planning with turns matched what the robots actually did more closely and shortened the total time to finish compared with classical planning, and experiments with TurtleBot4 robots showed that it works in practice.



