Our research focuses on algorithms for robots that learn from human feedback efficiently, quantify their own uncertainty honestly, and coordinate with human partners fluently.
Robots deployed in the real world face distribution shift, novel tasks, and shifting user preferences — and the most reliable source of correction is the human in the loop. We develop algorithms that let robots learn continually from rich human feedback: demonstrations, corrections, and preferences — robust to teachers whose attention is intermittent, whose feedback is noisy, and whose goals change mid-deployment.
Representative work: Optimal Interactive Learning on the Job via Facility Location Planning (RSS 2025); Conformalized Interactive Imitation Learning (ICLR 2025); LLMs for Preference-Based Sequence Prediction (ICAART 2025).
How does a black-box robot policy know when it doesn't know? We use conformal prediction to give robot policies calibrated confidence over their own decisions, so they can flag risky states before acting on them, defer to a human when out of their depth, and ask for help in a way that preserves task progress rather than interrupts it.
Representative work: Conformalized Teleoperation (RSS 2025); Conformalized Interactive Imitation Learning (ICLR 2025).
How does a person know how to use a robot when they don't know what it is capable of? We develop methods for robots to explain their high-level strategies, behavioral preferences, and uncertainty to human teammates — making collaboration legible and predictable, and making human feedback actually helpful.
Representative work: Multi-Agent Strategy Explanations for Human-Robot Collaboration (ICRA 2024); Bi-Directional Mental Model Reconciliation for HRI with LLMs.
When a robot is already deployed, every query to a human teammate has a cost: it interrupts the task, fragments the user's attention, and consumes time. We design algorithms that decide when and what to ask in deployment, framing query selection as a planning problem that balances progress across a multitask sequence, the marginal value of new information, and the teacher's limited bandwidth.
Representative work: Optimal Interactive Learning on the Job via Facility Location Planning (RSS 2025).
Effective teammates model each other and adapt. We study how robots can infer a partner's strategy from a handful of observed actions, communicate their own intentions, and adjust their behavior to support fluent joint action — across collaborative manipulation, shared autonomy, and assistive robotics.
Representative work: Learning Human Contribution Preferences (CoRL 2023); Coordination with Humans via Strategy Matching (IROS 2022); The Role of Adaptation in Human-AI Teaming (topiCS).
Teaching a robot isn't easy. People differ in how they perform a task and in how well they can teleoperate, demonstrate, or give feedback. We build learning systems that are robust to this heterogeneity — separating beneficial corrective data from harmful data, and detecting when new input is misaligned with what a policy has already learned.
Representative work: Conformalized Teleoperation: Confidently Mapping Human Inputs to High-Dimensional Robot Actions (RSS 2025).
Grounding robot failure prediction in causal concept bottleneck models, so a robot can say which concepts drove a predicted failure and what would change the outcome. Presented at the Human-Robot Dialogue workshop at IROS 2026.
Analyzing what a policy's internal visual representations encode and turning them into stage-relevant guidance for human teachers during interactive corrections. Presented at the Rethinking Uncertainty workshop at IROS 2026.
Understanding users' needs, preferences, and interactions with robots, and designing interfaces that let non-experts personalize and shape robot behavior.
Bringing user-centered design methods and immersive technologies to interfaces for observing, teaching, and collaborating with robots.
User research and strategic design for human-in-the-loop robot learning: studying how people actually intervene, correct, and collaborate with learning robots to inform interaction design.
See the publications page for the full list.
RIA Lab