Interview with Haimin Hu: Sport-theoretic integration of security, interplay and studying for human-centered autonomy

On this interview collection, we’re assembly among the AAAI/SIGAI Doctoral Consortium contributors to search out out extra about their analysis. On this newest interview, Haimin Hu tells us about his analysis on the algorithmic foundations of human-centered autonomy and his plans for future tasks, and provides some recommendation for PhD college students seeking to take the subsequent step of their profession.

May you give us an outline of the analysis you carried out throughout your PhD?

My PhD analysis, carried out below the supervision of Professor Jaime Fernández Fisac within the Princeton Secure Robotics Lab, focuses on the algorithmic foundations of human-centered autonomy. By integrating dynamic sport concept with machine studying and safety-critical management, my work goals to make sure autonomous techniques, from self-driving autos to drones and quadrupedal robots, are performant, verifiable, and reliable when deployed in human-populated house. The core precept of my PhD analysis is to plan robots’ movement within the joint house of each bodily and knowledge states, actively making certain security as they navigate unsure, altering environments and work together with people. Its key contribution is a unified algorithmic framework—backed by sport concept—that permits robots to securely work together with their human friends, adapt to human preferences and targets, and even assist people refine their abilities. Particularly, my PhD work contributes to the next areas in human-centered autonomy and multi-agent techniques:

• Reliable human–robotic interplay: Planning secure and environment friendly robotic trajectories by closing the computation loop between bodily human-robot interplay and runtime studying that reduces the robotic’s uncertainty in regards to the human.
• Verifiable neural security evaluation for advanced robotic techniques: Studying strong neural controllers for robots with high-dimensional dynamics; guaranteeing their training-time convergence and deployment-time security.
• Scalable interactive planning below uncertainty: Synthesizing game-theoretic management insurance policies for advanced and unsure human–robotic techniques at scale.

Was there a venture (or facet of your analysis) that was significantly attention-grabbing?

Security in human-robot interplay is particularly tough to outline, as a result of it hinges on an, I’d say, nearly unanswerable query: How secure is secure sufficient when people would possibly behave in arbitrary methods? To offer a concrete instance: Is it adequate if an autonomous car can keep away from hitting a fallen bicycle owner 99.9% of the time? What if this charge can solely be achieved by the car at all times stopping and ready for the human to maneuver out of the best way?

I might argue that, for reliable deployment of robots in human-populated house, we have to complement normal statistical strategies with clear-cut strong security assurances below a vetted set of operation circumstances as properly established as these of bridges, energy crops, and elevators. We want runtime studying to attenuate the robotic’s efficiency loss brought on by safety-enforcing maneuvers; this requires algorithms that may scale back the robotic’s inherent uncertainty induced by its human friends, for instance, their intent (does a human driver wish to merge, minimize behind, or keep within the lane?) or response (if the robotic comes nearer, how will the human react?). We have to shut the loop between the robotic’s studying and decision-making in order that it might optimize effectivity by anticipating how its ongoing interplay with the human might have an effect on the evolving uncertainty, and finally, its long-term efficiency.

What made you wish to research AI, and the world of human-centered robotic techniques particularly?

I’ve been fascinated by robotics and clever techniques since childhood, once I’d spend total days watching sci-fi anime like Cellular Go well with Gundam, Neon Genesis Evangelion, or Future GPX Cyber Components. What captivated me wasn’t simply the futuristic know-how, however the imaginative and prescient of AI as a real associate—augmenting human talents reasonably than changing them. Cyber Components particularly planted the thought of human-AI co-evolution in my thoughts: an AI co-pilot that not solely helps a human driver navigate high-speed, high-stakes environments, but additionally adapts to the motive force’s fashion over time, finally making the human a greater racer and deepening mutual belief alongside the best way. At present, throughout my collaboration with Toyota Analysis Institute (TRI), I work on human-centered robotics techniques that embody this precept: designing AI techniques that collaborate with individuals in dynamic, safety-critical settings by quickly aligning with human intent by way of multimodal inputs, from bodily help to visible cues and language suggestions, bringing to life the very concepts that when lived in my childhood creativeness.

You’ve landed a school place at Johns Hopkins College (JHU) – congratulations! May you speak a bit in regards to the technique of job looking out, and maybe share some recommendation and insights for PhD college students who could also be at an analogous stage of their profession?

The job search was undoubtedly intense but additionally deeply rewarding. My recommendation to PhD college students: begin considering early in regards to the sort of long-term impression you wish to make, and act early in your software bundle and job speak. Additionally, be sure you speak to individuals, particularly your senior colleagues and friends on the job market. I personally benefited lots from the next sources:

Do you might have an thought of the analysis tasks you’ll be engaged on at JHU?

I want to assist create a future the place people can unquestionably embrace the presence of robots round them. In direction of this imaginative and prescient, my lab at JHU will examine the next matters:

• Uncertainty-aware interactive movement planning: How can robots plan secure and environment friendly movement by accounting for his or her evolving uncertainty, in addition to their capacity to cut back it by way of future interplay, sensing, communication, and studying?
• Human–AI co-evolution and co-adaptation: How can embodied AI techniques study from human teammates whereas serving to them refine present abilities and purchase new ones in a secure, personalised method?
• Secure human-compatible autonomy: How can autonomous techniques guarantee prescribed security whereas remaining aligned with human values and attuned to human cognitive limitations?
• Scalable and generalizable strategic decision-making: How can multi-robot techniques make secure, coordinated selections in dynamic, human-populated environments?

How was the expertise attending the AAAI Doctoral Consortium?

I had the privilege of attending the 2025 AAAI Doctoral Consortium, and it was an extremely worthwhile expertise. I’m particularly grateful to the organizers for curating such a considerate and supportive surroundings for early-career researchers. The spotlight for me was the mentoring session with Dr Ming Yin (postdoc at Princeton, now college at Georgia Tech CSE), whose insights on navigating the unsure and aggressive job market have been each encouraging and eye-opening.

May you inform us an attention-grabbing (non-AI associated) reality about you?

I’m keen about snowboarding. I discovered to ski primarily by vision-based imitation studying from a chairlift, although I’m undoubtedly paying the value now for poor generalization! At some point, I hope to construct an exoskeleton that teaches me to ski higher whereas conserving me secure on the double black diamonds.

About Haimin

Haimin Hu is an incoming Assistant Professor of Pc Science at Johns Hopkins College, the place he’s additionally a member of the Knowledge Science and AI Institute, the Institute for Assured Autonomy, and the Laboratory for Computational Sensing and Robotics. His analysis focuses on the algorithmic foundations of human-centered autonomy. He has acquired a number of awards and recognitions, together with a 2025 Robotics: Science and Methods Pioneer, a 2025 Cyber-Bodily Methods Rising Star, and a 2024 Human-Robotic Interplay Pioneer. Moreover, he has served as an Affiliate Editor for IEEE Robotics and Automation Letters since his fourth 12 months as a PhD pupil. He obtained a PhD in Electrical and Pc Engineering from Princeton College in 2025, an MSE in Electrical Engineering from the College of Pennsylvania in 2020, and a BE in Digital and Data Engineering from ShanghaiTech College in 2018.