In tests using the RoboCasa GR-1 benchmark, extending the visual context from 0.0 to 19.2 seconds increased the task success rate from 63.3% to 78.7%. The findings indicate that having access to historical data is insufficient unless the model is specifically designed to apply that history to its current actions.
Tuesday, October 20, 2026
Story
Nvidia's Long-WAM Raises Robot Success Rate to 78.7% on RoboCasa
AI12 days ago
The chipmaker released a new AI framework that enables robots to better navigate tasks by expanding the amount of video data they can reference during execution. This Long-WAM world-action model scales visual context for real-time control, optimizing the way models utilize historical visual data when the video foundation is pretrained autoregressively.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.