Spring Robotics Colloquium: Tapomayukh Bhattacharjee (Cornell)
Tapomayukh Bhattacharjee, who leads the EmPRISE Lab at Cornell University, discusses building personalized physical caregiving robots, drawing on work with users, caregivers, and occupational therapists at Cornell and UW.
In this talk
- 3:53
Caregiving depends on the person and setting
Bhattacharjee describes caregiving as a mix of tasks, individual abilities and preferences, environments, and caregiver approaches. For feeding, for example, one person may move their neck while another signals readiness by looking at a caregiver, so the same task calls for different strategies. He argues that robots need to serve a broad range of people while adapting to each user.
- 12:25
Feeding robots need to balance performance and user effort
The feeding system combines food recognition and task planning for bite acquisition with sensing and control for delivering food safely. Because failures remain common, the team developed QueryGap to weigh the robot’s confidence against the workload of asking a user for help. Data from more than 100 people informed its workload model, and a limited user study found that accounting for user factors improved satisfaction.
- 32:36
Home deployment reveals the need for adaptation and safety
In a five-day home deployment, a study of around 20 people with mobility limitations produced 46 distinct feeding preferences; the team implemented 36. Flexible interfaces, user-specific safety checks, and a fork designed to fold under a measured force limit helped address varied needs. The robot also explained unexpected behavior, such as waiting too long after interpreting a user’s instruction.
- 40:33
Personalizing contact across different caregiving tasks
For bed bathing, RAG-Diff uses stored observations and actions to adapt a diffusion policy to explicit preferences, such as areas to avoid or force limits. Separate work explores whole-arm contact using wireless force-sensitive resistor sensors, and whole-body transfers using FlexADT, a kit that instruments wheelchairs, Hoyer lifts, and hospital beds. The bathing trials used a mannequin, and the transfer system had not yet undergone user studies.
- 54:14
A practical recipe for caregiving robots
Bhattacharjee’s approach combines user-aware representations, modular skills, simulation, flexible interfaces, and clear explanations of robot behavior. He describes a deployment where changing the fork’s orientation to match a user’s preference made the feeding task possible. The goal is to support caregivers when they are present and help meet needs when they are not.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.