CMRDF: Morphology-Conditioned Reachability Fields for Task-Space Robot Policy Transfer

Generalist robot policies often represent actions as end-effector target poses in task space, which facilitates policy transfer across robots performing the same task. However, a task-space pose that is feasible for one morphology may be infeasible for another due to joint limits, workspace differences, or self-collision constraints. Existing pose reachability methods often learn monolithic fields over the full SE(3) pose space, which spreads expensive inverse-kinematics supervision over a high-dimensional domain and leaves near-boundary regions poorly sampled in cross-morphology settings. We propose Cross-Morphology Reachability Distance Fields (CMRDF), a morphology-conditioned task-space reachability representation that decomposes full-pose feasibility into global position reachability and position-conditioned orientation reachability. The two outputs are signed-distance function (SDF) proxies with physical units. CMRDF uses dense forward-kinematics coverage and distance-to-boundary targets for the Position-SDF, and inverse-kinematics boundary mining at representative position anchors for the Orientation-SDF. The differentiable fields support reachability queries and gradient-based local pose repair. Under matched full-pose inverse-kinematics budgets, CMRDF improves SE(3) accuracy over a capacity-matched monolithic baseline by 2.5 percentage points at both 1.00× and 0.51×. It also supports low-distortion pose repair and zero-shot transfer under joint-limit and same-family geometry shifts. In target-side policy fine-tuning, converted demonstrations reduce the required native data from 60–80 to 20 demonstrations at 90% of a target-only reference policy's success rate.