Shivendra Agrawal

About me: I am a postdoctoral researcher at Yale University in the Interactive Machines Group, working with Prof. Marynel Vázquez on in-the-wild human-robot interaction and robot legibility, and part of the NSF Science and Technology Center for Human and Robot Co-Adaptation. I completed my Ph.D. in Computer Science at CU Boulder, advised by Prof. Brad Hayes, where I developed deployable, context-aware robotics and human-centered AI at the intersection of robotics, HRI, and vision-language models. My work has appeared at venues including RA-L, IROS, HRI, AAMAS, and ICRA.

Research vision: I build robots and AI systems that understand and adapt to people in real-world spaces like homes, stores, and hospitals, by modeling the implicit context (semantic, social, and physical cues) that lab settings leave out. My research spans three threads: context-aware embodied AI, assistive and accessible robotics, and human-robot and human-AI co-adaptation, including how AI tools can strengthen people’s own skills rather than replace them.

Teaching and mentoring: Great teachers drew me into computer science, and I try to pay that forward. I have taught or TA’d six courses at CU Boulder, earning multiple teaching awards, and mentored eight undergraduate and M.S. researchers. More on my teaching and mentorship pages.

Selected research:

Embodied AI

Multimodal Knowledge Extraction and Spatial Grounding (In Submission): Turns mobile LiDAR scans into a lightweight semantic map that grounds vision-language models in 3D space, enabling intent-aware search, localization, and language-guided navigation. Project page

Embodied AIAssistive Robotics

Robust Semantic Localization (RA-L): A semantic particle filter for accurate global localization in cluttered, changing environments using only low-cost RGB-D and visual-inertial sensors. Project page

Assistive RoboticsCo-Adaptation

Learning Decision-Making Policies from Human Behavior (AAMAS): A robotic cane that learns a guidance policy from human behavior to help blind users find and grasp products on store shelves (featured in national media). Project page

Assistive Robotics

Designing Embodied AI for Social Context (IROS): A perceptive robotic cane that uses models from psychology to guide users to socially appropriate seats, respecting privacy and personal space. Project page

Co-Adaptation

Explainable Robotic Coaching (HRI): A robot infers a person's mental model from their mistakes and coaches them with explanations (HRI '19 Best Paper Runner-up). Project page

Other interests: Outside of research, I enjoy playing tennis, biking, running, and exploring local food.