NEXUS LAB
Nexus Lab connects human judgment with AI execution, so your team can turn ideas into shippable results faster.
NEXUS LAB
Build policy reuse on a verifiable protocol across robot embodiments.
Router SDK defines observations, actions, capabilities, and feedback. Adapters handle device differences so models can focus on the task and integration work can be reused across devices and projects.
Shared observation and action definitions connect data collection, standard datasets, and training adapters, creating a common foundation for policy training and deployment.
Distinguish native, convertible, and unsupported actions. Coordinate transforms, inverse kinematics, and scheduling help match requests to each robot’s capabilities.
Native / Convertible / Unsupported
Coding Agent helps generate, complete, and debug adapter code using vendor documentation and existing implementations, reducing repetitive integration work.
Specify · Generate · Validate
Check capabilities and constraints, then capture execution status and errors. Preserve adapters, configurations, and validation records for future devices and projects.
Validate → Execute → Feedback
OUR FRAMEWORK
Router SDK, runtime management, and embodiment adapters connect applications to vendor interfaces through shared observations, actions, capabilities, and feedback. A specification and adapter library supports Coding Agent in generating and debugging reusable adapters.
SIMULATION & EVALUATION
Switch simulation environments to explore task-by-task comparisons.
All values are fictional and demonstrate the interface only. They are not NEXUS LAB results or official evaluations from these simulation environments.
Sample everyday manipulation tasks: picking, moving, and using drawers.
| Task | Baseline (sample) | Unified adaptation (sample) |
|---|---|---|
| Pick up a can | 62% | 82% |
| Move a cube | 68% | 86% |
| Open a drawer | 54% | 76% |
| Place in a drawer | 48% | 72% |
| Task average | 58% | 79% |
ONE MODEL, DIFFERENT DEGREES OF FREEDOM
Execution results for the same manipulation task, using one policy model adapted to robotic arms with different degrees of freedom. Slots are reserved for 6-DoF and 7-DoF configurations; demo footage is coming soon.
Reserved for a video of the shared manipulation task running on a 6-DoF arm.
Reserved for the same task on a 7-DoF arm, showing how actions adapt to a different number of joints.
SIMULATION DEMOS
Explore observation inputs, action execution, and task feedback in simulation as the model is adapted to arms with different degrees of freedom.
Explore simulated task scenes, target objects, and model observation inputs.
See simulated execution after model actions are mapped to arms with different degrees of freedom.
Explore task outcomes, runtime status, and execution feedback in simulation.
Simulation demo videos will be added here.