SUGARv2: Learning Diverse Visual Humanoid Loco-Manipulation Skills from Synthesized Interaction References
SUGARv2: Learning Diverse Visual Humanoid Loco-Manipulation Skills from Synthesized Interaction References
This research presents SUGARv2, a framework for acquiring diverse visual humanoid loco-manipulation skills by synthesizing interaction references and using a shared training pipeline. The authors adapt reusable task templates to different spatial configurations to generate coordinated robot and object motions at low cost. By employing reference-based imitation as a common learning objective, they train privileged teachers first, then distill visual policies through reinforcement learning with mimic rewards, leveraging diverse object instances and visual domain randomization for generalization and sim-to-real transfer. Across five humanoid tasks, SUGARv2 achieves a 91.0% mean success rate in simulation and shows successful sim-to-real transfer on a real robot for whole-body manipulation.
The paper introduces SUGARv2, a unified framework that learns diverse visual humanoid loco-manipulation skills by synthesizing interaction references and using reference-based imitation for training; it reports a mean simulation success rate of 91.0% across five tasks and demonstrates sim-to-real transfer on a real robot, though deployment readiness and safety are not independently verified.
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.
- Environment
- SimulationOpen source S4
Reported numbers
Reported success rate
91%
Reported trials
View original evidence
rough teacher distillation and reinforcement learning with the same reference-based mimic rewards, using diverse object instances and visual domain randomization to support visual generalization and sim-to-real transfer. Across five humanoid loco-manipulation tasks, SUGARv2 achieves a mean success rate of 91.0% in simulation. Deployment on a real humanoid robot further demonstrates successful sim-to-real transfer for
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
rough teacher distillation and reinforcement learning with the same reference-based mimic rewards, using diverse object instances and visual domain randomization to support visual generalization and sim-to-real transfer. Across five humanoid loco-manipulation tasks, SUGARv2 achieves a mean success rate of 91.0% in simulation. Deployment on a real humanoid robot further demonstrates successful sim-to-real transfer for
Open source S4
arXiv:2611.09266v1 Announce Type: new Abstract: Scaling up visual humanoid loco-manipulation skill acquisition requires reducing the human effort needed as tasks and spatial configurations expand, particularly in demonstration collection and task-specific training design. We present SUGARv2, a unified framework for learning diverse visual humanoid loco-manipulation skills through interaction reference synthesis and a
Open source S2
shared training pipeline. To obtain diverse references at low cost, we adapt reusable task templates to different spatial configurations, synthesizing coordinated robot and object motions. To reduce task-specific reward design and training effort, we use reference-based imitation as a common learning objective across tasks. We first train privileged teachers to execute these references, then train visual policies th
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org