Madi Dian, Ph.D.

The group develops physical AI for human-robot teaming. Current work includes conversational vision-language-action models that turn ordinary spoken instructions into safe mobile manipulation, multimodal intent decoding from EEG, EMG, eye tracking, and speech, shared autonomy that reads whether a person has actually committed to an action before the robot moves, haptics in teleoperation, digital twins for robot learning, and pipelines that convert ordinary human video into demonstration data in a robot's own embodiment.

A parallel line studies how robotics, mechatronics, and artificial intelligence should be taught, and produces open course materials and interactive companions used well beyond one classroom.

This work draws on robotics, machine learning, control theory, artificial intelligence, computer vision, digital twins and game development, psychology, physical therapy, agriculture, and education.

• Conversational mobile manipulation. Vision-language-action models that turn ordinary spoken instructions into safe mobile manipulation, and ask a clarifying question when an instruction is ambiguous instead of guessing.
• Intent recognition. Transformer models over EEG, EMG, eye tracking, and speech that separate real movement intent from imagined intent.
• Shared autonomy. Commit-readiness models that judge whether a person has settled on an action before the robot commits to it.
• Learning from human video.Pipelines that re-render recorded human demonstrations in a robot's own embodiment, so policies trained on them move the way people do.
• Digital twins and haptic teleoperation. Simulated testbeds where a policy is rehearsed before it reaches hardware, and force feedback that gives a remote operator the feel of contact.
• Robotics and AI education.Open course material on screw-theory kinematics, modern robotics, and mechatronic system design, with interactive companions used well beyond one classroom.

• A team player who is positive, creative, organized, and principle-centered.
• A solid foundation in robotics: kinematics, dynamics, control, and perception.
• Proficiency in at least one programming language, Python preferred, otherwise C++ or MATLAB.
• Familiarity with the Robot Operating System (ROS).
• Familiarity with machine learning, computer vision, natural language processing, and AI algorithms for intelligent and adaptive robotic systems.
• Experience integrating sensors such as cameras, depth sensors, and force sensors, and processing the resulting data.
• Knowledge of software development principles, software architecture, and version control.
• A strong research record shown by publications, together with analytical thinking and skill in experiment design and data analysis.
• Photography and videography skills are a plus.

Fracture-directed steerable needles are a class of needle that reaches a target by first creating a fracture in the soft tissue and then following it with a superelastic needle, with applications in surgery and biopsy.

The work produced kinematic models based on screw theory, a waterjet steering system, a finite-element and closed-form model of cut depth in soft tissue, and a patent, Fracture-directed steerable needles (US 11,103,278).