Odyssey's Agora-2 trains on Diablo II gameplay for real-time world sim with up to 20 humans and agents
testingcatalog · x · 2026-09-25
Odyssey (Oliver Cameron) released a playable research preview of Agora-2, a multi-agent world model supporting up to 20 humans and agents in a shared, real-time generated environment — 5x more participants than Agora-1, now across multiple environments with longer-horizon behaviors. It was trained on Diablo II captures pairing observations with player inputs and game state, functioning as a learned game engine; one model tracks each player's state while a second renders per-player views. The preview pits humans against AI bots trained to hunt them, and Odyssey says such worlds could train robots and security agents.
More from Embodied
- LeWAM: Lightweight World Action Model Hits 92.28% Success on RoboTwin 2.0 — udmrzn · 2026-09-25
- Blind Stargardt patient reads a letter again after PRIMA implant in Science clinical trial — lukaszkaiser · 2026-09-25
- Magnetic quadruped robot welds and inspects shipyard hulls, climbing vertical steel walls — LexiLove · 2026-09-25
- AGIBOT and Chimelong open world's first embodied-intelligence theme park with 300+ humanoids — robleclerc · 2026-09-25
- NeurIPS paper learns compositional motor options with adapter banks, cutting generalization error up to 10x — sreejan_kumar · 2026-09-25
- Nature Machine Intelligence: Cybernetics and interoception framework for embodied AI agents — AnnaCiaunica · 2026-09-25