ABot-World-0: Infinite Interactive World Rollout on a Single Desktop GPU
ABot-World-0 claims real-time, long-horizon interactive video rollout at 720P on one RTX 5090.
The paper describes an action-conditioned world model trained from AAA games, simulation engines, and internet video. It uses raw keyboard actions for control, plus a reference-character memory to keep third-person character identity consistent. The authors report up to 16 FPS, 1.2 seconds action-to-first-frame latency, and about 19GiB peak VRAM in optimized low-bit configurations. ArXiv · AI/CL/LG's note
The paper describes an action-conditioned world model trained from AAA games, simulation engines, and internet video. It uses raw keyboard actions for control, plus a reference-character memory to keep third-person character identity consistent. The authors report up to 16 FPS, 1.2 seconds action-to-first-frame latency, and about 19GiB peak VRAM in optimized low-bit configurations. ArXiv · AI/CL/LG's note
score 6