Human-Robot Mutual Adaptation in Shared Autonomy
TL;DR A joystick-driven assistive arm infers both the user's goal and how willing they are to adapt, then nudges adaptable users toward a better goal and gives in to stubborn ones. In a 51-person study, 75% of users in this condition switched to the better goal at least once, and they trusted the robot more than one that always ignored them.
What they do
Shared-autonomy assistance usually assumes the user’s goal is right. Here the robot knows a better goal and steers the user there only if they seem willing to follow; otherwise it complies to keep their trust. The test task: users drive an arm with a joystick (left, forward, right) to pick one of two bottles.
How it works
- Human model. A Bounded-memory Adaptation Model: the user reacts only to the last k robot actions and switches to the robot’s apparent goal with probability alpha (their adaptability), weighted by how sure they are of what the robot is doing.
- Partial observability. Both the user’s goal (mode) and alpha are hidden variables in a mixed-observability MDP. Each goal is a stochastic modal policy over paths, which tolerates noisy joystick input.
- Disagreement penalty. Since only the robot’s actions move the world, the reward adds a cost C for following a mode different from the user’s. Against a stubborn user this cost outweighs the better goal’s reward.
- Solver. 520 states (5 alpha values x 2 modes x 52 observable states), k=1, R_opt=11, R_sub=10, C=-0.32, solved offline with SARSOP.
Results
- In 1000 simulated runs, the more adaptable the user, the more often the robot reached the optimal goal.
- Human study, n=51, between subjects. Mutual adaptation: 9/16 users reached the optimal goal in the final trial and 12/16 in at least one trial. One-way adaptation (always assist the user’s goal): only 2/17 changed goals.
- Trust (3.94/5) and perceived collaboration (3.91/5) under mutual adaptation were higher than with a robot that always went to the optimal goal (p=0.010, p=0.017), and not significantly different from one-way adaptation.
Limitations
One two-goal task with three discrete actions, and the “optimal” goal was set to whichever bottle the participant did not pick. The robot is assumed to know the optimal goal, and adaptability is assumed fixed during the task. Observable states grow as the number of robot actions to the power k, so larger action sets or longer memory would need feature-based approximations.
Figures
Abstract
Shared autonomy integrates user input with robot autonomy in order to control a robot and help the user to complete a task. Our work aims to improve the performance of such a human-robot team: the robot tries to guide the human towards an effective strategy, sometimes against the human's own preference, while still retaining his trust. We achieve this through a principled human-robot mutual adaptation formalism. We integrate a bounded-memory adaptation model of the human into a partially observable stochastic decision model, which enables the robot to adapt to an adaptable human. When the human is adaptable, the robot guides the human towards a good strategy, maybe unknown to the human in advance. When the human is stubborn and not adaptable, the robot complies with the human's preference in order to retain their trust. In the shared autonomy setting, unlike many other common human-robot collaboration settings, only the robot actions can change the physical state of the world, and the human and robot goals are not fully observable. We address these challenges and show in a human subject experiment that the proposed mutual adaptation formalism improves human-robot team performance, while retaining a high level of user trust in the robot, compared to the common approach of having the robot strictly following participants' preference.
Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.