Shen Li

= Ph.D. candidate at MIT, advised by Prof. Julie A. Shah.
∈ Interactive Robotics Group, CSAIL.
Research goal = enable robots to:
    - safely interact with humans
    - efficiently accomplish collaborative tasks.

Previously,
- M.S. in Robotics at CMU,
     Co-advised by Prof. Siddhartha Srinivasa + Prof. Stephanie Rosenthal.
     ∈ Personal Robotics Lab (now at UW), Robotics Institute.
- Internship at CMU, advised by Prof. Katia Sycara.
- B.S. in Computer Science + Psychology at Penn State.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

• 12/2022: Our work on imitation learning (CoRL'22 Oral) was presented at CoRL 2022.
• 09/2022: We installed 3 interactive exhibitions at MIT Museum: 1. interact, 2. teach task, 3. teach motion.
• 06/2022: Our work on robot dressing (RAL'22) was interviewed by Voice Of America (from 12:07).
• 04/2022: Invited talk at Prof. Dana Kulic's lab at Monash University.
• 04/2022: Our work on state estimation for robot dressing (RAL'22) appeared in MIT news and Instagram.
• 01/2022: Our work on robot-assisted dressing (RSS'21) was featured in Next Byte Podcast with article.
• 10/2021: I received NeurIPS 2021 Outstanding Reviewer Award (top 8% of reviewers).
• 09/2021: Invited talk at Open Learning's MIT Horizon: Safe Human-Robot Interaction in Personal Care.
• 07/2021: Our work on safe robot dressing (RSS'21) appeared in MIT CSAIL news and Instagram.
• 12/2019: Demonstrated human-robot comm. (HRI'20) at the MIT AeroAstro holiday party (photos).

Set-based state estimation (1) with a probabilistic consistency guarantee, when dynamic and observation models are learned via GPs; (2) that can be formally reduced to its probabilistic counterpart - GP-EKF; robot-assisted dressing.

Human-aware motion planning with (1) a probabilistic guarantee on human physical safety, and (2) an improved task efficiency thanks to a two-pronged definition of safety: collision avoidance OR safe impact; robot-assisted dressing.

Robot motion-planning for safe (via collision avoidance) and efficient (via planning in both space-time) human-robot collaboration.

Task-and-motion planning under robust against a human operator's cooperative or adversarial interventions.

Robot decision-making on if, when, and what to communicate during collaboration.

Semi-supervised human model learning => low efforts on parameter specification & high performance on robot collaborative decision-making.

Fast online human activity recognition + robot assistance at the appropriate time.

Trust and performance in supervisory control of swarm robots with varied levels of autonomy.

Imitation learning for continuous motion + discrete plan (as LTL formula).

Robot decision-making under uncertain & non-Markovian task specifications.

Robot learning non-Markovian task specifications from human demonstrations.

Robot learning non-Markovian task specifications from human demonstrations.

A dataset of natural language instructions for object specification in manipulation scenarios (1582 instructions from online crowdsourcing).

Effectiveness of critical way-points in conveying info about robot objectives.

Correlations between clarity of spatial reference instructions and 1) perspective taking 2) spatial features.

Webpage design courtesy of Jon Barron
Accessibility