Ground Robot MRTA: Auction Baselines vs. MARL

Multi-robot task allocation (MRTA) is where the "many small units cooperating" theme becomes concrete: a team of resource-constrained ground robots needs to divide up a set of tasks with no single robot able to see the whole picture.

Baseline: auction algorithms

The baseline allocator uses a distributed auction algorithm — robots bid on tasks based on local cost estimates, and allocations are refined through repeated rounds of bidding.

Learned coordination

A MARL policy is trained against the same task distribution to see whether learned coordination can close the gap to a centralized optimum, and under what conditions the auction baseline is simply good enough.