Wearable robots may improve running training efficiency and reduce injury risk, yet almost every existing device augments gait through joint torques delivered at the hip or ankle. Whether an external propulsive force from a body-worn thruster—an actuation principle that injects linear momentum into the whole body rather than augmenting a single joint—can instead assist fast gait has not been tested, and it is unknown whether neuromuscular control can exploit such a force without losing balance. Existing evidence cannot resolve this question: joint-torque exosuits did not test the alternative actuation principle; many simulation-based designs optimized a tracked trajectory that cannot change in response to a device; and applying a force to a fixed, never-assisted controller measures only an out-of-distribution perturbation. This study therefore introduced a reinforcement learning (RL) co-adaptation framework in an 80-muscle, 34-degree-of-freedom musculoskeletal forward-dynamics simulation. The neuromuscular controller was warm-started from a 150-million-step pretrained policy and re-optimized for each assistance condition, so the gait was free to adapt to the device. Across four controlled comparisons, co-adaptation proved methodologically necessary: the gait of a fixed controller degraded under a 10 N force and fell outright at 100 N, and training from scratch failed on the 80-muscle model. Under the same framework, idealized hip-flexion assistance reduced the cost of transport (COT) by 27–41% across two complementary metabolic models, whereas a body-worn forward thruster at 20 N or 50 N did not converge to a durable beneficial gait. In a representative 50 N bout, the thruster drove a divergent forward trunk lean (a pitching instability; 23^∘ → 47^∘ →70^∘ ) that collapsed the gait at ~1.8\,s . The site of assistance, not only its magnitude, was decisive: in the tested simulation, body-worn propulsive assistance for fast gait constituted a postural-stability problem rather than an energetic one.
No takes yet. Share an insight, caveat, or question.
Yan et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: