An integrated pipeline using five IMUs and a QDTPformer to recognize terrain and select task-specific DMP references for powered knee-exoskeleton assistance achieved 95.74% offline accuracy and an equally weighted mean sEMG RMS reduction of 8.42%, with task- and wearer-dependent responses, in six participants.
QDTPformer recognized three terrain tasks (Flat, Ramp, Stairs) with 95.74% offline test accuracy.
Under leave-one-subject-out evaluation, QDTPformer reached 89.54%±1.07% mean window-level accuracy.
Online recognition selected the corresponding DMP reference and assistance parameters for the powered knee exoskeleton.
Phase-aligned DMP-knee correlations were high for Flat and Ramp and lower for Stairs.
Assistance reduced sEMG RMS by a mean of 8.42% across two muscles and three tasks, with responses varying by task and wearer.
The study is framed as a feasibility evaluation whose main contribution is an end-to-end pipeline assessed with both kinematic and muscle-activity outcomes across three terrains.
This study tested a wearable knee exoskeleton that figures out what kind of terrain the wearer is walking on and adjusts its assistance accordingly. Five motion sensors on the body fed into an AI model (a Transformer) that distinguished level walking, walking up a ramp, and climbing stairs. The system then chose a matching assistance pattern for the knee motor while the person carried a load. The terrain-recognition model was about 96% accurate on test data, and about 90% accurate when tested on people it had not seen during training. The assistance pattern matched the wearer's knee movement very closely on flat ground and ramps, and less closely on stairs. On average, muscle activity in two leg muscles fell by about 8% across the three tasks, but the effect differed between tasks and between wearers. This research suggests that combining terrain recognition with task-specific assistance is feasible. However, only six people took part, so the results are preliminary. The modest and uneven muscle reductions, and the weaker match on stairs, suggest that more testing and refinement would be needed before drawing firm conclusions about real-world benefit.
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Upload Your LabsLiu J, Lin Y, Yao Y, Zhu Y, Zhang H, Wang X, et al.. (2026). Integrated Multi-Terrain Locomotion Recognition and Knee-Exoskeleton Assistance for Load Carrying: A Feasibility Study.. Biosensors. https://doi.org/10.3390/bios16100570