LessMimic: Versatile Humanoid-Object Interaction with Unified Distance Field Representations
LessMimic: Versatile Humanoid-Object Interaction with Unified Distance Field Representations
This research presents LessMimic, a framework enabling versatile whole-body interaction for humanoid robots with everyday objects without relying on motion-reference inputs at inference. The system uses a single whole-body policy steered by root commands and an interaction-type flag, conditioned on compact representations derived from distance fields and local surface directions. Through visual distillation, it supports egocentric-depth-based deployment without motion capture. The framework maintains robust performance across object scales from 0.4x to 1.6x on tasks including PickUp, SitStand, Push, and Carry, while supporting sequential composition of heterogeneous skills with 62.1% success on five-task sequences and 23.5% at 15 instances.
The paper introduces LessMimic, a framework for humanoid-object interaction that uses unified distance field representations to enable geometry-conditioned control without motion-reference inputs at inference, demonstrating robust performance across object scales and sequential task compositions.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
object scale range
0.4
View original evidence
from short histories of distance-field-derived surface distances and local surface directions. This geometry-conditioned interface enables command-steered control while adapting contact behavior to local object geometry. Through visual distillation, LessMimic further enables egocentric-depth-based deployment for MoCap-free sensing. Across object scales from 0.4x to 1.6x, LessMimic maintains robust performance over fo
Open source S4object scale range
1.6
View original evidence
from short histories of distance-field-derived surface distances and local surface directions. This geometry-conditioned interface enables command-steered control while adapting contact behavior to local object geometry. Through visual distillation, LessMimic further enables egocentric-depth-based deployment for MoCap-free sensing. Across object scales from 0.4x to 1.6x, LessMimic maintains robust performance over fo
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ur interaction tasks, including PickUp, SitStand, Push, and Carry, where motion-conditioned baselines degrade sharply away from the training scale. Beyond single-task evaluation, LessMimic further supports sequential composition of heterogeneous interaction skills, attaining 62.1% success on five-task sequences and retaining 23.5% success at 15 task instances. By grounding interaction control in local geometry rather
Open source S5
from short histories of distance-field-derived surface distances and local surface directions. This geometry-conditioned interface enables command-steered control while adapting contact behavior to local object geometry. Through visual distillation, LessMimic further enables egocentric-depth-based deployment for MoCap-free sensing. Across object scales from 0.4x to 1.6x, LessMimic maintains robust performance over fo
Open source S4
arXiv:2602.21723v2 Announce Type: replace Abstract: Humanoid robots capable of versatile whole-body interaction with everyday objects represent a central goal of embodied intelligence. Existing approaches often rely on motion-reference inputs or task-specific rewards, coupling policies to particular motion scripts, object geometries, and contact timings. This limits geometric generalization and flexible interaction c
Open source S2
ontrol, where simple commands specify motion intent while object geometry determines how contact should be executed. We introduce LessMimic, a framework for versatile motion-free-at-inference humanoid-object interaction. At inference, a single whole-body policy is steered by root commands and an interaction-type flag instead of motion-reference inputs, and is conditioned on a compact interaction representation built
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org