IsaacLab users flag missing wrist wrench observation vs. Gazebo in AI challenge
Developers ask if IsaacLab needs a force-torque sensor equivalent...
A developer participating in the AI for Industry Challenge raised a critical discrepancy between IsaacLab and Gazebo simulation environments. While Gazebo exposes a wrist wrench observation via a force-torque sensor broadcaster, IsaacLab does not offer an equivalent built-in observation. The user, Senne_Bogaerts, is exploring IsaacLab to run further experiments but cannot find a direct wrench observation in the current API.
The developer asks if querying wrenches or contact forces at specific frames—like the wrist3 frame or ATI/tool frame—is the intended alternative, or if a proper force-torque sensor is missing and should be added. They emphasize that without this observation, experiments in IsaacLab cannot closely match the Gazebo evaluation setup, potentially affecting reproducibility and fair competition. The post also references related topics, indicating broader interest in aligning environments. This issue underscores a need for clearer documentation or feature parity between simulation tools used in robotics challenges.
- IsaacLab lacks a dedicated wrist wrench observation that Gazebo provides via a force-torque sensor broadcaster.
- Users suggest querying wrenches at specific frames (wrist3, ATI/tool) as a workaround but seek official guidance.
- The discrepancy impacts the AI for Industry Challenge by making experiments in IsaacLab less comparable to Gazebo-based evaluations.
Why It Matters
Simulation fidelity affects robot learning; missing sensor equivalences can skew benchmark results and hinder cross-platform reproducibility.