RoboBoom
Rice Ken Kennedy Institute· March 23, 2021· 54:51

Dr. Lydia Kavraki and Dr. Angela Radulescu on Robotics and Human-computer Interaction DS C2C Seminar

Rice University robotics researcher Lydia Kavraki and NYU Center for Data Science cognitive scientist Angela Radulescu discuss robot motion planning and active data collection, alongside human attention and decision-making, at a Data Science Coast to Coast Seminar hosted by seven university data science institutes.

In this talk

  1. 7:58

    Robots as embodied data collectors

    Kavraki describes a robot as an embodied system that senses its surroundings, reasons about them, plans, and acts. Examples include mapping a building with a laser rangefinder and reconstructing a room from an RGB-D sensor’s point cloud. Robots can also gather ongoing measurements such as temperature, noise, air quality, and Wi-Fi strength, then seek out missing information themselves.

  2. 10:56

    Planning motion in challenging environments

    Kavraki explains how sampling-based planners explore possible robot positions and connect them into roadmaps or trees, balancing exploration with efficient paths. Her group worked with Eduard Vidal’s team at ICM in Spain on an underwater vehicle planner that adapts as sonar reveals obstacles while respecting dynamics and safety constraints. For NASA’s Robonaut, the team planned constrained movements for tasks such as retrieving cargo bags and pushing a station door.

  3. 16:48

    Learning from tasks and human interaction

    Kavraki describes storing noisy task data so a robot can retrieve useful experience and plan faster when it encounters similar work. In one example, a robot moves cans into a bin and pushes it based on a high-level mission, without exact object positions or a fixed action order. When a person removes a block from an arch the robot is building, the robot must recognize the disruption and recover the task.

  4. 24:32

    Attention shapes learning and search

    Radulescu studies how people represent tasks in natural settings and how those processes may inform mental health research. In a repeated-choice study, eye tracking and neuroscience-based measures showed that attention affected both participants’ choices and how they learned from feedback; gaze patterns fit an inference account better than gradual trial-and-error learning. In virtual reality, participants found targets in cluttered rooms 75% of the time, and models using object color and shape produced search patterns similar to people’s.

  5. 43:41

    Mood, attention, and future therapies

    Radulescu describes models in which mood tracks recent differences between expected and received rewards. Simulations showed that attention to different features can change modeled mood even when outcomes stay the same, and an experiment found that happy and sad moods biased attention toward positive and negative expectations, respectively. She sees potential for closed-loop therapies that guide attention toward features associated with healthier mood.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks