Walking with a support robot should not feel like operating a machine. A person does not naturally communicate every step, turn or moment of instability through buttons or spoken commands. Instead, movement intent and the need for support are continuously expressed through posture, motion and physical contact.
A research team from Zhejiang University, Tencent Robotics X Laboratory and Shanghai Jiao Tong University has developed a walking-support robot that uses large-area tactile contact at the human torso as its main channel for physical interaction. The study was published online in National Science Review recently.
The work was inspired by hands-on walking assistance from a physical therapist. A therapist can guide a person at the waist while simultaneously sensing posture, motion and changes in balance through contact. The researchers translated this principle into a robotic system that maintains compliant contact at mechanically meaningful regions around the waist and underarms.
Large-area tactile interfaces embedded in the robot's two arms measure not only how much force is applied, but also where the contact occurs and how pressure is distributed. A forward or backward shift of the contact region reflects the user's movement tendency. An asymmetric distribution between the two sides creates a turning signal. These changing tactile patterns directly modulate the robot's motion, rather than first being converted into discrete labels such as “move forward” or “turn left.”
The tactile interface is integrated with whole-body state estimation and a multi-layer control framework. A centroidal dynamics model represents the user's center-of-mass motion and angular momentum. Nonlinear model predictive control plans suitable contact locations and support forces, while tactile admittance control continuously adjusts the planned motion according to real-time contact. A constraint-aware quadratic programming controller then executes the motion while enforcing robot joint limits.
This architecture allows intention alignment and balance support to emerge from the same physical channel. The robot does not need to switch between separate modes for normal walking, turning and fall response. It continuously adapts as the coupled human-robot dynamics evolve.
In experiments, users walked forward and backward and turned left and right without issuing explicit commands. The spatial distribution of tactile contact changed systematically with each movement, and the robot adjusted its base and arm motion smoothly. During sudden lateral balance disturbances, the robot first repositioned its arms toward a more mechanically effective support region and then increased the supporting force. As the user recovered, the force gradually decreased and the system returned to normal walking assistance without a separate fall-detection trigger.
The researchers also evaluated physiological and subjective responses in six healthy adult participants under three conditions: normal walking without robotic support, handle-based robotic assistance and torso-level tactile assistance. Hamstring surface electromyography showed a significant overall effect of walking condition. Torso-level assistance produced lower muscle activity than handle-based assistance in all six participants, with a statistically significant paired comparison. Five of the six participants also showed lower activity than during normal walking, although this comparison did not reach conventional statistical significance.
In a post-trial questionnaire, participants rated torso-level assistance more highly than handle-based assistance for perceived support and comfort. They also reported higher willingness to use the torso-level system in daily life. Compared with normal walking, torso-level assistance received higher ratings for support, safety and perceived reduction in physical effort.
The study suggests that tactile sensing can play a deeper role than simply detecting contact. By coupling the robot to the torso over a broad area, touch becomes a bidirectional physical communication channel: the user continuously influences the robot's motion, while the robot continuously redistributes contact and support in response.
The researchers caution that the present experiments were conducted in a controlled environment with healthy adults. The current framework is intended for mild-to-moderate balance disturbances in which the user's feet remain planted. Larger studies involving older adults, people with gait impairments and clinical populations will be needed, together with additional recovery strategies for severe disturbances that require stepping.
The findings provide a path toward assistive robots that act less like command-driven devices and more like physically cooperative partners. Rather than waiting for an instruction or an emergency event, the robot can remain continuously engaged with the user's movement through touch.
National Science Review
Experimental study