2024 Fall Robotics Colloquium: Elaine Short (Tufts)

In this talk

1. 1:30

   Robotics is a design problem

   Short shows how robots turn images, sensor readings, and other inputs into numerical representations and actions. She argues that those representations and system choices are design decisions with consequences for people. For example, a delivery robot waiting at a curb cut can block a wheelchair user from reaching the sidewalk, undermining access created through disability advocacy.

2. 8:12

   Design with disability in mind

   Short describes her Assistive Agent Behavior and Learning Lab and its use of human-centered design and disability community ethics. She frames disability as part of human variation and locates many barriers in inaccessible environments rather than in disabled people. Accessibility, flexibility, interdependence, and care can inform robotics for everyone.

3. 15:52

   Use human input to shape robot learning

   Short explains how reinforcement learning maps states to actions, while noting that a robot can learn behavior that earns reward yet still fails to match what people want. Her group studies feedback, demonstrations, and scalar ratings, including a method that turns ratings into binary feedback with confidence. In long-horizon tasks such as entering a PIN, demonstrations improved learning when used to train the world model instead of directly imitating each human action.

4. 38:55

   Account for how people teach and control

   Short’s group studied how human feedback differs from idealized oracle feedback in delay, frequency, strictness, bias, and accuracy. In user studies, actual teaching differed from participants’ expectations, and the severity and type of an error affected how people taught. In shared-control work, users adapted to incorrect inferred goals; a painting system with adjustable parameters offered a middle ground between direct control and choosing from preset autonomous agents.

5. 55:09

   Make robotics spaces more accessible

   Short highlights research by Kat Allen on barriers faced by disabled makers and makerspace operators. The team found a cycle in which inaccessible spaces discourage disabled participation, reducing pressure to improve access, and developed recommendations. She also points students to mentoring and support through Tapia, CRA-WP, and AccessComputing.