Explainability Subcard
| Field | Response |
|---|---|
| Intended Task/Domain: | Task-specific humanoid robot policy for bimanual apple handover and plate placement, in simulation. |
| Model Type: | Robot VLA model, fine-tuned from NVIDIA GR00T N1.7 Base. |
| Intended Users: | Developers, researchers and community working on robot foundation models. |
| Output: | 61-D upper-body joint targets and base velocity in the GR00T evaluation route. |
| Describe how the model works: | Accepts vision, language and proprioception, then outputs a robot action policy. |
| List the groups (or group characteristics) for which this was tested to produce comparable outcomes: | Not Applicable |
| Technical Limitations & Mitigation: | This model is not tested or intended for use in mission-critical applications that require functional safety. Use in such applications is at the user's own risk and sole responsibility, including taking the necessary steps to add appropriate guardrails, safety monitoring, collision avoidance, and validation procedures prior to deployment. This model was trained and evaluated entirely in simulation and has not been validated on physical hardware. More generally, limitations include, but are not limited to: - The model may underperform in environments, object configurations, or task variations that differ from the training distribution. - Performance may degrade under challenging visual conditions, including poor lighting, shadows, occlusions, sensor noise, or camera viewpoints not represented during training. - Performance can vary across robot embodiments, grippers, sensor configurations, and control interfaces; additional fine-tuning and validation may be required for new deployments. Risks and possible mitigations include: Risk: Model underperformance in environments with different object placements, lighting conditions, camera configurations, or scene layouts. Mitigation: Expand training data coverage, incorporate domain randomization, and fine-tune the model using data representative of the target deployment environment. Risk: Performance degradation when transferring from simulation to physical robots. Mitigation: Perform embodiment-specific fine-tuning, sim-to-real adaptation, and validation on representative real-world tasks prior to deployment. Risk: Integration challenges across robot embodiments with different kinematics, end-effectors, sensors, or control interfaces. Mitigation: Provide embodiment-specific integration guidance and validate performance on target robot platforms. Risk: Unstable grasps, object drops, collisions, or task failures when operating outside the validated task distribution. Mitigation: Deploy external safety mechanisms, collision checking, grasp validation, human supervision, and runtime monitoring systems. |
| Verified to have met prescribed NVIDIA quality standards: | Yes |
| Performance Metrics: | Success rate (369/500, 73.8%, in closed-loop simulation evaluation), as well as the following: 1) if the trajectory is smooth and does not jitter 2) if the robot does not hit any other objects 3) if the trajectory is natural |
| Potential Known Risks: | This model is not tested or intended for use in mission critical applications that require functional safety. It has been evaluated only in simulation and must not be deployed to physical hardware without independent safety and behavior qualification. The use of the model in those applications is at the user's own risk and sole responsibility, including taking the necessary steps to add needed guardrails or safety mechanisms prior to deployment. |
| Licensing: | NVIDIA Open Model License (https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) |