Robotics · 2026-10-16
Dexterity from Supervised Play
Botao He, Yiming Dou, Yiannis Aloimonos, Jitendra Malik, Akash Sharma, Ruoshi Liu
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
DSP (Dexterity from Supervised Play) trains a goal-agnostic play policy by behavior cloning on minutes of human play, then uses it to autonomously collect hours of interaction for an action-conditioned world model. At inference, the play policy proposes candidates for MPC under the learned model to achieve arbitrary goals; optionally a closed-loop goal-conditioned policy is distilled via self-play in the world model. Evaluated in simulation and on real hardware.
Why it matters
The authors propose a framework that turns supervised, goal-free human play into (1) autonomous data collection for training a world model and (2) goal-directed manipulation via model-predictive control using the play policy as an action prior. The paper reports real-world results where the method outperforms hindsight behavior cloning and achieves comparable success to reinforcement learning with far less interaction data, and it claims no task-specific demonstrations or sim-to-real transfer are needed.
Method
- Train a goal-agnostic play policy by behavior cloning on goal-free human teleoperation data; safe behavior is not explicitly engineered but implicitly demonstrated.
- Roll out the play policy autonomously to collect observation-action transitions, then train an action-conditioned world model to predict future observations (Diffusion Forcing).
- At inference, use play-guided MPC: sample candidate action chunks, score predicted outcomes by goal reward, execute selected actions; optionally distill the MPC closed loop via world-model self-play.
Abstract (from arXiv)
How can robots autonomously collect the contact-rich experience needed to learn dexterous manipulation? We present DSP: Dexterity from supervised play, a framework in which a robot learns a play policy from minutes of human play through behavior cloning, then deploys it autonomously to collect hours of interaction data to train a world model. During inference, the same play policy then guides model-predictive control (MPC) by proposing candidate action sequences for evaluation under the learned world model. To accelerate inference, we also introduce a method for distilling a closed-loop, goal-conditioned policy through self-play within the learned world model. We evaluate this framework on underactuated pushing and any-to-any in-hand pose reorientation with a multifingered hand, in simulation and real world. In real world experiment, our method outperforms hindsight behavior cloning by 69.6% in success rate and achieves comparable success to reinforcement learning with 490 times less interaction data. These results demonstrate that supervised play can autonomously collect interaction data, guide model-based planning, and enable policy learning without any task-specific demonstrations or sim-to-real transfer.
Related papers
- ForceWAM: Decoupling Future and Response Learning from Demo and Play Data for Contact-Rich Manipulation
- Direct Experience World-Model Optimization: Learning the World Beyond Action Imitation
- DexForesight: Grounding Dexterous Policies with Implicit Spatial Foresight
- Perceive, Predict, and Plan: a Unified World Action Model for Egocentric Dexterous Manipulation
- PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-15 · Indexed by Pub-AI: 2026-10-16 UTC
Dexterity from Supervised Play
Preprint posted: 2026-10-15 · Original record ↗
- arxiv
- 2610.22340
Explicit source versions (1)
- v1 · 2026-10-15
- Preprint available
Source date: 2026-10-15 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
How can robots autonomously collect the contact-rich experience needed to learn dexterous manipulation? We present DSP: Dexterity from supervised play, a framework in which a robot learns a play policy from minutes of human play through behavior cloning, then deploys it autonomously to collect hours of interaction data to train a world model. During inference, the same play policy then guides model-predictive control (MPC) by proposing candidate action sequences for evaluation under the learned world model. To accelerate inference, we also introduce a method for distilling a closed-loop, goal-conditioned policy through self-play within the learned world model. We evaluate this framework on underactuated pushing and any-to-any in-hand pose reorientation with a multifingered hand, in simulation and real world. In real world experiment, our method outperforms hindsight behavior cloning by 69.6% in success rate and achieves comparable success to reinforcement learning with 490 times less interaction data. These results demonstrate that supervised play can autonomously collect interaction data, guide model-based planning, and enable policy learning without any task-specific demonstrations or sim-to-real transfer.
