cs.ROOct 13, 2026

ForceWAM: Decoupling Future and Response Learning from Demo and Play Data for Contact-Rich Manipulation

Authors: Zijun Xu, Wendi Chen, Xinkai Wang, Jiachen Zou, Hongjin Chen, Hairuo Liu, Bokai Lin, Shihao Ma, 

Organizations: Fudan University · Shanghai Innovation Institute · Shanghai Jiao Tong University · Southeast University · 

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Expert demonstrations show how to complete contact-rich tasks, but provide limited coverage of how interaction forces respond to different actions. Task-agnostic play provides broader experience of contact dynamics without specifying task goals. To fully leverage these two types of data, we introduce ForceWAM, a force world action model that decouples knowledge of task-compatible interaction outcomes from knowledge of how local actions change contact. A world model learns from successful demonstrations to jointly predict future video and six-axis force/torque trajectories. Together, these futures describe the expected evolution of motion and contact during the task. A separate inverse dynamics model (IDM) learns from both task and play, mapping current observations and recorded visual-wrench futures to their associated robot actions. Play data thus provide knowledge of force responses for goal-directed execution. At deployment, the IDM reactively translates generated futures into actions, reusing them with fresh wrench feedback between world-model updates. Across real-robot force-regulated sliding, box flipping with and without disturbances, and ring placement, ForceWAM achieves 8585-95%95\% mean success, surpassing world-modeling and reactive baselines. Our experiments further show that adding task-agnostic play data improves success by 1515-4545 percentage points, demonstrating its value for learning contact-rich execution. Project page and code are available at https://force-wam.github.io/.

Figures & tables

Figure 1: Motivation of ForceWAM. The illustrated end-to-end force-conditioned and force-envisioned policies use only task demonstration data: the former lack pretrained video-world-model priors, while the latter lack within-chunk force feedback. ForceWAM combines the pretrained prior with task demonstrations for task-compatible visual-wrench futures and action-labeled play for IDM execution learning, achieving the highest mean success across all four real-robot conditions.
Figure 2: ForceWAM training and deployment. Task demonstrations train the visual-wrench future generator, while task and play data train the IDM from recorded futures and associated actions. During deployment, ForceWM supplies future conditions and the causal IDM updates actions using measured wrench feedback.
Figure 3: Real-robot comparisons of ForceWAM, RDP, ForceVLA, and FA-WAM.
MethodSlide  ↑\,\uparrowFlip  ↑\,\uparrowFlip-D  ↑\,\uparrowHanoi  ↑\,\uparrow
RDP ( 2025 )15%40%30%45%
ImplicitRDP ( 2026 )20%45%25%55%
ImplicitRDP ( 2026 ) ++ Play10%25%10%30%
ForceVLA ( 2025 )15%55%30%60%
Cosmos Policy ( 2026 )15%40%30%60%
Cosmos Policy ( 2026 ) ++ Play10%20%15%40%
Table 1: Mean real-robot success rates (the best per column in bold). Slide requires full-path completion with 55 - 15 15\, N contact force. “ ++ Play”: training with additional action-labeled play data.
Figure 4: Mean success with increasing play data. Task demos, the ForceWM model, and IDM optimizer steps are fixed.
Figure 5: ForceWM predictions during closed-loop box flipping. (a) Recorded wrist images (top) and decoded futures (bottom). The three central predictions share one conditioning observation. (b) Recorded (gray) and predicted (blue) FyF_{y} and FzF_{z} (all six wrench components in Appendix F ).
Future conditioningSlide  ↑\,\uparrowFlip  ↑\,\uparrowFlip-D  ↑\,\uparrowHanoi  ↑\,\uparrow
Visual ++ wrench85%95%85%90%
Visual only15%40%30%70%
Wrench only20%30%25%30%
No future conditioning15%30%35%50%
Visual ++ wrench, no augmentation65%80%70%75%
Table 2: Online future-conditioning ablation: mean success rates. All IDMs use task ++ play data, with futures from one fixed ForceWM. Rows 1-4 vary future modalities during training and deployment. The last uses both without training-time future-conditioning augmentation.
Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: Hardware setup and task objects.
AugmentationFlip-box setting
Whole-condition dropout0.100.10
Visual / wrench dropout0.250.25 / 0.500.50
Scale jitter[0.85,1.15][0.85,1.15]
Wrench noise std.0.0750.075 in normalized coordinates
Visual-latent noise std.0.0750.075 times latent standard deviation
Temporal offsets{0,−1,−2,−3,−4,−5,−6,−9,−12}\{0,-1,-2,-3,-4,-5,-6,-9,-12\}
Appendix
Table 3: Future-conditioning augmentation used for the flip-box IDM.
Figure 7: Full six-axis comparison for the rollout in Figure 5 . Recorded (gray) and predicted (blue) forces are shown on the left and torques on the right.
More in this topic

Contact-Rich Robotic Manipulation

Contact-rich robotic manipulation studies how robots plan and control physical interactions such as grasping, insertion, pushing, and dexterous object handling.

See all 410 papers

Explore similar work

CardsList

  1. FAWAM: Force-Aware World Action Models for Closed-Loop Contact-Rich Manipulation

    Jun 7, 2026Haotian He, Zeyu Yan, Qipeng Liu +2Robotic ControlWorld Models for Robotics

    School of Mathematical Sciences, Peking University · School of Artificial Intelligence, Shanghai Jiao Tong University.

  2. TacWAM: Anchor-Guided World Action Model with Mechanics-Aware Tactile Prediction

    Jul 30, 2026Lei Jin, Yiding Ma, Xin Zhang +3World Models for RoboticsContact-Rich Robotic Manipulation

    1Tsinghua University · 2Manifold AI

  3. CW-WAM: A Structured Contact-Wrench World Action Model for Force-Sensitive Manipulation

    Oct 21, 2026Houzhu Ding, Xinda Qi, Luqi Wang +1Contact-Rich Robotic ManipulationWorld Action Models

    Ant Group