Overview
Cortex 2.0 extends our original Cortex architecture by introducing a world model into the learning loop. It uses a world model to predict a set of future scene rollouts in video latent space, helping the robot to evaluate potential futures and choose the most promising actions, thus avoiding costly failures and improving efficiency in robotic manipulation tasks.
Flagship features
- Cortex 1.6 improves success rate by 6–9% over Cortex 1.5 and by 15–25% over previous models
- Recovery success rates improved from ~45% (baseline) to ~80% (Cortex 1.6)
- Dense rewards allow for faster learning, with up to 3× quicker convergence time compared to the baseline
- Designed to continuously improve while performing tasks in real-world settings, utilizing feedback from operational data
Specifications
Category: Cobot- Type
- other
- Force limiting
- Yes
- F/T sensing
- none
- API / SDK
- No
Detailed specifications
Compute1
- Ros Compatible
- false
Other8
- Price Tier
- 150K+
- Applications
- pick_and_place,returns_handling,kitting
- Deployment Notes
- Cortex 2.0 has been deployed in various settings, including return handling, kitting, parcel closing, and item handling. Notable customer implementations include Active Ants, Deltilog, Arvato, and Radial.
- Industries Served
- warehouse,logistics
- Availability Status
- available
- Compatible Grippers
- Robotiq 2F-140
- Programming Interface
- drag_and_drop
- Additional Information
- - Cortex 1.6 improves success rate by 6–9% over Cortex 1.5 and by 15–25% over previous models. - Recovery success rates improved from ~45% (baseline) to ~80% (Cortex 1.6). - Dense rewards allow for faster learning, with up to 3× quicker convergence time compared to the baseline. - Designed to continuously improve while performing tasks in real-world settings, utilizing feedback from operational data.
Compare with peers
Reviews for Cortex 2.0
Loading reviews…
