Research Highlights
Progressive Learning for Physics-informed Neural Motion Planning
Abstract— A new semi-linear Eikonal PDE formulation and progressive speed scheduling strategy to solve high-DOF motion planning problems.
RSS'23NTFields: Neural Time Fields for Physics-Informed Robot Motion Planning
Abstract— Unlike most neural motion planners, NTFields require no expert trajectories for training and instead directly learn to solve Eikonal PDE.
ICLR'23 [SPOTLIGHT]CoGrasp: 6-DoF Grasp Generation for Human-Robot Collaboration
Abstract— Robot grasp generation contextualizes human social preferences of interacting with daily-life objects for human-robot collaboration.
ICRA'23Robot Active Neural Sensing and Planning
Abstract— A framework that actively collects the RGBD observations of an unknown confined environment with an in-hand camera and transforms them into interpretable scene representation.
TRO'23NeRP: Neural Rearrangement Planning for Unknown Objects
Abstract— NeRP is a learning-based approach for multi-step neural object rearrangement planning with never-before-seen objects in the real world.
RSS'21Constrained Motion Planning Networks X
Abstract— CoMPNetX is a neural planning approach with a fast projection operator for solving constrained manipulation tasks.
TRO'21