Accurate human torque estimation is crucial for enabling task-agnostic control in robotic exoskeleton systems. However, estimation errors may cause mismatches between the robot assistance and the human intention, degrading controllability and task performance. In this paper, we address this issue by formally defining matched assistance as scenarios in which the robot positively contributes to human movement. Based on this definition, we develop a theoretical framework to design the robot's desired interaction torque that guarantees a lower bound on the matched assistance probability. Importantly, the proposed guarantee holds over the entire torque distribution, including unseen data beyond the training tasks. This provides our method with strong reliability and generalization, both of which are critical for effective exoskeleton control. The proposed strategy is implemented on the ABLE upper-limb exoskeleton and evaluated in a multi-task setup. Experimental results validate the theoretical guarantees and demonstrate that the proposed strategy achieves effective general performance across several tasks, guaranteeing movement smoothness while reducing human physical effort.
Figures & tables
Fig. 1: Exoskeleton control strategy. Rather than directly passing the output of the high-level torque estimation model to the low-level controller as in [ 14 ] , we introduce a middle-level controller where the estimated torque is modulated by a dead-zone mechanism to ensure a high matched assistance probability before being applied to the low-level controller.
Fig. 2: Illustration of assistance behavior derived from the relationship between the human’s intended and the robot’s desired interaction torque under modulation approaches: direct assistance [ 14 ] (top plot), scaled assistance [ 15 ] (middle plot), and the proposed dead-zone-based assistance (bottom plot).
Fig. 3: Testing tasks: PTT = Pure trajectory tracking, LTT = Load trajectory tracking, RR = Reach and return, PP = Pick and place.
Fig. 4: Results on generalization performance guarantee across 17 subjects for the shoulder and elbow torques. The gray lines denote evaluated values from the same subject. (A) Torque estimation error of considered estimation models with corresponding generalization error upper bounds LB∗ . (B) Matched assistance probability P(Ma) of considered estimation models without the dead-zone (Tdz=0) , with the dead-zone (Tdz=2LB∗) ; the horizontal lines indicate the lower guarantee PB∗=0.89 given by ( 14 ).
Fig. 5: The distribution of estimation error ( τh−τ^h ) obtained from 5 considered models across 17 subjects. Each curve corresponds to one subject.
Fig. 6: Average PCI across 17 subjects and two joint torques derived by five considered torque estimation models: MVLR, NLMap, 4HNN, LSTM, and CNN-LSTM.
Fig. 7: Results on (A) motion smoothness and (B) muscle activation of 5 subjects across all testing tasks: PTT = Pure trajectory tracking, LTT = Load trajectory tracking, RR = Reach and return, PP = Pick and place. The last column, AVE, shows the average of the 5 subjects.
Fig. 8: Comparison between SCA and DZA across 17 subjects in terms of P(Ma) and average PCI.
Sep 14, 2026·Xiao-Yin Liu, Guotao Li, Weiqun Wang +1ExoskeletonTorque
State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China. · The School of Artificial Intelligence, University of Chinese Academy Sciences, Beijing, China.
The State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, China · The School of Artificial Intelligence, University of Chinese Academy of Sciences, China · The School of Automation and Intelligence, Beijing Jiaotong University, China +1