Research theme
Platforms and tools
Open-source simulators, benchmarking libraries, and custom hardware platforms — VMAS, BenchMARL, RoboMaster, Raven, and Sanity.
We design and maintain the open-source infrastructure that the multi-agent community builds on. VMAS is a vectorised simulator for large-scale multi-agent experiments; BenchMARL provides a standardized benchmarking library for reproducible MARL research. On the hardware side, the Cambridge RoboMaster gives us an agile ground platform, while our Raven and Sanity aerial platforms support indoor swarm experiments at lab scale.
Recent papers
- BenchMARL: Benchmarking Multi-Agent Reinforcement Learning JMLR · 2024
- The Cambridge RoboMaster: An Agile Multi-Robot Research Platform DARS 2024 · 2024
- POPGym: Benchmarking Partially Observable Reinforcement Learning ICLR 2023 · 2023
People
- Amanda Prorok
- Jan Blumenkamp
- Jennifer Gielis
- Manon Flageat
- Mateusz Odrowaz-Sypniewski
- Maxence Boels
- Sridhar Sola