Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory have demonstrated a system called RoboRaise that allows a robot to assist with lifting tasks by monitoring the muscle activity in a user’s arm, rather than through spoken commands or explicit programming.
The system places electromyography (EMG) sensors on a user’s biceps and triceps. According to MIT CSAIL, the sensors detect the firing of neurons as a person tenses or relaxes muscles, and the system’s algorithms continuously estimate changes to the person’s arm level. A neural network trained on data from previous users detects discrete up-and-down hand gestures for finer control.
Joseph DelPreto, a graduate student and lead author on the project, described the design intent in the CSAIL announcement: “Our approach to lifting objects with a robot aims to be intuitive and similar to how you might lift something with another person — roughly copying each other’s motions while inferring helpful adjustments. The key insight is to use nonverbal cues that encode instructions for how to coordinate, for example to lift a little higher or lower. Using muscle signals to communicate almost makes the robot an extension of yourself that you can fluidly control.”
DelPreto co-authored the paper with MIT Professor and CSAIL Director Daniela Rus, who framed the broader goal: “We aim to develop human-robot interaction where the robot adapts to the human, rather than the other way around. This way the robot becomes an intelligent tool for physical work.”
The team tested RoboRaise with 10 participants through three lifting experiments. In the first, the robot did not move. In the second, it moved in response to muscle signals but did not help lift the object. In the third, person and robot lifted together. MIT CSAIL reports that when users received feedback from the robot — either seeing it move or lifting an object with it — achieved heights were significantly more accurate than when no feedback was present.
Gesture detection worked correctly in roughly 70 percent of cases, according to the announcement. The team also ran assembly tasks, including lifting a rubber sheet onto a base structure, and reported successful handling of both rigid and flexible objects.
RoboRaise was implemented on the team’s Baxter humanoid robot, but MIT CSAIL says the system could be adapted to other robotic platforms.
A new user can start with minimal calibration: after putting on the sensors, they tense and relax their arm a few times and lift a light weight to a few heights. The gesture-detection neural network is trained only on data from previous users, not on per-session data from the new user.
The researchers note that EMG signals present practical challenges. They are often noisy, and estimating limb position from muscle activity alone is difficult. RoboRaise addresses this by keeping a human in continuous control rather than inferring intent autonomously. The use of wearable sensors also avoids problems that can affect vision- or speech-based systems, such as occlusions or background noise.
DelPreto told MIT CSAIL he could imagine RoboRaise being used in manufacturing, construction, and household assistance settings.
The project extends an earlier CSAIL system that allowed users to correct robot mistakes using brainwaves and hand gestures. That prior system focused on discrete corrections; RoboRaise enables continuous motion in a more collaborative mode.
Future work, according to the team, includes adding sensors for additional muscles to increase degrees of freedom, and incorporating cues like exertion or fatigue. The team also tested a version that uses biceps and triceps activity levels to communicate how stiffly the person is holding their end of an object, allowing the pair to either drag the object or hold it taut, depending on the signal.
The team presented their work at the International Conference on Robotics and Automation (ICRA).