Cortex 2.0: Grounding World Models in Real-World Industrial Deployment
Cortex 2.0 is an upgraded robotic manipulation system that shifts from simply reacting to situations towards a more strategic approach of planning and acting. It generates possible future paths in a visual latent space, evaluates them for their likelihood of success and efficiency, and picks the best option. This advancement addresses the challenges faced by Vision-Language-Action models during longer tasks. In trials involving single-arm and dual-arm setups across four different tasks—picking and placing items, sorting trash and items, sorting screws, and unpacking shoeboxes—Cortex 2.0 consistently outperformed top competitors.
Key facts
- Cortex 2.0 generates candidate future trajectories in visual latent space.
- It scores trajectories for expected success and efficiency.
- It commits only to the highest-scoring candidate.
- It shifts from reactive control to plan-and-act.
- Evaluated on single-arm and dual-arm manipulation platforms.
- Tasks include pick and place, item and trash sorting, screw sorting, and shoebox unpacking.
- Outperforms state-of-the-art Vision-Language-Action baselines.
- Designed for reliable long-horizon execution across embodiments, tasks, and changing object distributions.
Entities
—