Learning forceful manipulation remains a key challenge in robotics. This work introduces ForceBand, a low-cost, wearable sEMG wristband that estimates fingertip forces from muscle activity, combined with a force-aware policy learning approach that enables robots to execute contact-rich tasks by predicting both motion and force trajectories from human demonstrations.
Wearable Force Sensing Design
The sEMG band is designed to accurately capture wrist muscle activity relevant to fine-grained finger control. The design balances sensing accuracy, low cost, and reproducibility, targeting two specific goals.
Accessibility: Low Cost and Easy Manufacturing. The ForceBand hardware is kept low-cost and easy to reproduce by relying on commodity components and standard prototyping workflows. The mechanical structure can be fabricated with common tools such as a commercial 3D printer, while the electronics are modular and sourced from readily available parts. The total cost can be as low as $300 depending on supplier. The complete bill of materials is open-sourced.
Performance: Muscle-aware and High Precision. The band collects high-quality data by combining muscle-aware electrode placement with a bipolar design. This performance is supported by three hardware choices: low-noise biopotential acquisition, bipolar differential sensing, and anatomically guided electrode placement.
Muscle-Aware sEMG Wristband Design
The sEMG module design is optimized for recording subtle electrical potentials at the wrist through three design choices: a high-quality medical biopotential measurement chip, a bipolar electrode design, and muscle-aware electrode placement.
Force-Aware Policy Learning from Human Demonstrations
To convert force-enriched human demonstrations into robot-executable policies, the approach extends a flow matching transformer policy so that force enters as an additional channel on both the observation side and the action side. The pipeline takes synchronized human video and estimated fingertip forces as input, retargets human hand motion to a robot embodiment, and trains a flow-matching policy to predict both motion and force trajectories. This allows the robot to learn not only where to move but also how much force to apply during contact-rich manipulation.
Ablation on Electrode Placement
For the 1-, 2-, and 4-channel settings, the released dataset is used with unused channels masked out. To compare even placement with the muscle-aware layout, two separate 30-minute datasets were collected, one for each placement. Both use the same protocol: 10 objects, 2-, 3-, and 5-finger grasps, repeated 5 times per object.
Increasing the number of channels consistently reduces force prediction error. Under the matched 30-minute protocol, the muscle-aware placement further improves over the evenly spaced 8-channel layout by 18%. These results show that anatomically guided placement provides more informative EMG signals than uniform wrist coverage.
Note: The left and right sub-tables should not be directly compared as they are using different data.
Limitations and Discussion
Absolute force accuracy. Force prediction from sEMG is not as accurate as direct force sensing, but its current accuracy is sufficient for many forceful manipulation tasks. The force prior learned from human data helps the policy understand when and how hard to apply force, and residual errors can be further reduced with limited robot interaction data or reinforcement learning using the robot's own force feedback.
Cross-user generalization. Muscle signals vary across users, so the current pipeline involves 4 users and applies per-user calibration. This limitation may be mitigated by scaling data collection: large-scale sEMG datasets have shown strong cross-user generalization, and related wrist sEMG controllers now operate across users without calibration. Larger EMG2Force datasets are expected to reduce or remove the need for user-specific calibration.
Calibration still requires fingertip force sensors. Calibration currently relies on fingertip force sensors for ground truth, so setup is not yet fully sensor-free. This requirement may be reduced by scaling EMG2Force to a larger and more diverse user population, or by developing a sensor-free calibration protocol based on standardized grasp interactions with calibrated objects or fixtures.
Frequently Asked Questions
What makes the ForceBand design different from existing sEMG wristbands? The ForceBand uses muscle-aware electrode placement informed by wrist anatomy, combined with a bipolar differential sensing design, achieving 18% lower force prediction error compared to evenly spaced electrode layouts.
How does the robot learn force-based manipulation from human demonstrations? Human video and estimated fingertip forces are simultaneously recorded, then a flow-matching transformer policy is trained to predict both motion and force trajectories, enabling the robot to learn contact-rich tasks with appropriate force application.
What are the main limitations of using sEMG for force prediction in robotics? Absolute force accuracy is lower than direct force sensing, cross-user generalization requires per-user calibration, and the calibration process still depends on fingertip force sensors for ground truth data.
Can the ForceBand work without per-user calibration? Not yet with the current dataset, but the authors expect that larger EMG2Force datasets could reduce or eliminate the need for user-specific calibration, as seen in related commercial wrist sEMG controllers.