# Contact-Rich Manipulation

4 related articles

Subscribe to updates

arXiv Oct 14, 2026
ForceWAM splits contact learning in two: demos set the goal, play data teach force response, reaching 95% on one real-robot task
In contact-rich manipulation, expert demonstrations cover successful paths but say little about how force responds to different actions, while task-agnostic play records broader contact dynamics without a task goal. ForceWAM gives the task-compatible motion and six-axis force/torque future to a force world model trained on demonstrations, and gives the question of which actions realize that future to an inverse dynamics model trained on both demonstrations and play; at deployment, fresh wrench feedback corrects actions between world-model updates. Across force-regulated sliding, box flipping with and without disturbances, and ring placement, the full system reaches 85%, 95%, 85%, and 90% success, above the listed baselines.
In contact-rich manipulation, expert demonstrations cover successful paths but say little about how force responds to different actions, while task-agnostic play records broader contact dynamics without a task goal. ForceWAM gives the task-compatible motion and six-axis force/torque future to a force world model trained on demonstrations, and gives the question of which actions realize that future to an inverse dynamics model trained on both demonstrations and play; at deployment, fresh wrench feedback corrects actions between world-model updates. Across force-regulated sliding, box flipping with and without disturbances, and ring placement, the full system reaches 85%, 95%, 85%, and 90% success, above the listed baselines.
Fudan University Shanghai Innovation Institute Shanghai Jiao Tong University Southern University of Science and Technology Southeast University
Zijun Xu（First author）
ForceWAM World Action Model Contact-Rich Manipulation
arXiv Oct 14, 2026
Robot wiping learns force, not just motion: one demo becomes a contact prompt, 78.7% on unseen objects
Encoding motion and fingertip contact forces from a single demonstration into an ordered prompt lets a frozen policy run wiping, rotation, and insertion on unseen objects at 78.7% overall success, 13.0 percentage points above the strongest baseline.
聂显（First author）
Contact Trajectory Prompting In-Context Robot Learning Contact-Rich Manipulation Flow Matching
arXiv Oct 12, 2026
One policy, one pass/fail signal: all 20 NIST assembly tasks without per-task training
AlphaNIST trains a single reinforcement-learning policy across all 20 assembly tasks on the NIST Assembly Task Board 1 using only a binary success reward, with no demonstrations, progress shaping, scripted primitives, or per-task tuning. Prior learning-based methods typically solved one task at a time or relied on fixed sequences, staged initial poses, and dense reward guidance. In simulation, the shared policy reaches 98.2% overall training success and outperforms separately trained specialists on small-part tasks.
AlphaNIST trains a single reinforcement-learning policy across all 20 assembly tasks on the NIST Assembly Task Board 1 using only a binary success reward, with no demonstrations, progress shaping, scripted primitives, or per-task tuning. Prior learning-based methods typically solved one task at a time or relied on fixed sequences, staged initial poses, and dense reward guidance. In simulation, the shared policy reaches 98.2% overall training success and outperforms separately trained specialists on small-part tasks.
Octi Zhang（First author）
AlphaNIST NIST Assembly Task Board #1 Multi-Task Reinforcement Learning Contact-Rich Manipulation Reinforcement Learning
arXiv Oct 5, 2026
RATE lifts UniVTAC eight-task macro-average success to 71.8% and, with 0.59% of parameters updated, raises both success and contact-safe success on three real-world tasks
The authors propose Risk-Aware Tactile Encoding (RATE): bilateral tactile history is aggregated by a two-layer LSTM into interaction context, shaped jointly by short-horizon future contact-change prediction and continuous task-conditioned alert supervision, then added to a frozen pretrained ACT policy through a zero-initialized residual adapter that updates about 0.59% of parameters; RATE reaches 71.8% macro-average success on the eight UniVTAC simulation tasks (retrained ACT+UniVTAC: 47.4%) and, on real-world USB Insertion, Bottle Cap Screwing, and Plug Insertion, achieves success rates of 75%, 85%, and 60% with contact-safe success rates of 93.33%, 100%, and 83.33%.
The authors propose Risk-Aware Tactile Encoding (RATE): bilateral tactile history is aggregated by a two-layer LSTM into interaction context, shaped jointly by short-horizon future contact-change prediction and continuous task-conditioned alert supervision, then added to a frozen pretrained ACT policy through a zero-initialized residual adapter that updates about 0.59% of parameters; RATE reaches 71.8% macro-average success on the eight UniVTAC simulation tasks (retrained ACT+UniVTAC: 47.4%) and, on real-world USB Insertion, Bottle Cap Screwing, and Plug Insertion, achieves success rates of 75%, 85%, and 60% with contact-safe success rates of 93.33%, 100%, and 83.33%.
Yuyao Jiang（First author）
RATE Tactile Representation Learning Alert Supervision Lightweight Residual Adaptation Contact-Rich Manipulation

You’re all caught up

ForceWAM splits contact learning in two: demos set the goal, play data teach force response, reaching 95% on one real-robot task

Robot wiping learns force, not just motion: one demo becomes a contact prompt, 78.7% on unseen objects

One policy, one pass/fail signal: all 20 NIST assembly tasks without per-task training

RATE lifts UniVTAC eight-task macro-average success to 71.8% and, with 0.59% of parameters updated, raises both success and contact-safe success on three real-world tasks