Spring 2021 GRASP SFI: Zackory Erickson - April 7th, 2021
Zackory Erickson, a PhD candidate in Georgia Tech’s Healthcare Robotics Lab, discusses robotic caregiving and physical human-robot interaction with the GRASP Lab audience.
In this talk
- 4:46
Predicting forces during dressing assistance
Erickson describes a recurrent model that uses force, torque, and motion readings from a robot’s end effector to estimate where clothing presses on a person’s arm. A separate prediction model and model predictive control let the robot anticipate contact and adjust its movement. In tests with a PR2, longer prediction horizons helped the robot avoid catching a hospital gown on participants’ elbows and hands.
- 18:34
Sensing limbs with capacitive sensors
To track people when cameras are blocked by the robot or clothing, Erickson presents capacitive sensors that detect limbs without touching them. A six-electrode sensor estimated four aspects of a limb’s relative pose, including distance and orientation, and supported robot movement during dressing and bathing tasks. Tests included four able-bodied participants; the approach also adapted to involuntary arm movement in a participant with cerebral palsy, though wet cloth made sensing less reliable.
- 30:19
Assistive Gym for simulation and benchmarks
Erickson introduces Assistive Gym, an open-source physics simulation framework for studying robots that provide physical assistance. It includes tasks such as bathing, feeding, and dressing, plus baseline controllers and benchmarks across robot-task combinations. Simulation also makes it possible to study human movement, body variation, impairments, and preferences at a scale that would be difficult to achieve with real participants.
- 40:27
Using virtual reality to improve transfer
Assistive VR Gym brings real people into simulations so researchers can assess virtual caregivers without physical robot contact. Early trials showed that policies trained in the original simulation did not work well with participants, in part because simulated body shapes and waist movements differed from theirs. After Erickson’s team adjusted the simulated biomechanics, improved policies performed better in a study with eight participants, including itch-scratching assistance in 62% of trials.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.