Daniela Rus

Distributed, networked, and self-organizing robots

In distributed and multi-robot systems, Rus has contributed algorithms and systems for coordination, coverage, motion planning, control, and collaboration among multiple machines. This work asks how many machines, each with limited local information, can coordinate to achieve a common goal. ^[26]

Her group has explored self-reconfiguring modular robots, swarm and collective robotic systems, modular robotic cubes, aerial and ground robot teams, and robotic systems that coordinate through local sensing and communication. In the Roboat and FloatForm projects, small autonomous boats sense, navigate, coordinate, connect, and self-assemble into larger floating structures such as temporary bridges and platforms. The work frames networked robotics as a problem of self-organization: how independent machines can become a coherent collective system. ^[27][28]

Soft, modular, and bio-inspired robotics

Rus has been a pioneer and an early contributor to soft robotics and bio-inspired robot systems. Her group has investigated robots made from compliant and unconventional materials, including silicone, paper, and edible materials, as alternatives to rigid industrial machines. Representative systems include ingestible origami robots for potential medical tasks inside the body, soft robotic fish for underwater observation, soft manipulators, modular robotic systems, and robots inspired by animal movement. ^[29]

This work has contributed to a broader definition of robots as machines that may be soft, small, foldable, modular, underwater, medical, or environmentally embedded, rather than necessarily humanoid or industrial in form. The underlying research question is how morphology, materials, sensing, and control together determine a machine's intelligence and capability. ^[30][31]

Computational design and fabrication

Rus has also worked on computational design and fabrication, including methods that combine algorithms, materials, manufacturing processes, sensing, and control to generate customized robots and functional devices. Her group has studied robot compilers, printable robots, programmable materials, sensorized structures, and methods for task-driven co-optimization of robot bodies and controllers. ^[32][33]

This research treats robot design as a computational problem: how software can help specify not only a robot's behavior, but also aspects of its body, actuation, sensing, materials, and fabrication. It connects robotics to design automation, computational fabrication, materials science, and AI-assisted engineering. ^[34]

Human-robot interaction and assistive systems

Rus's work in human-robot interaction examines how people can communicate with, teach, correct, and collaborate with intelligent machines. Her group has studied interfaces based on demonstrations, gestures, wearable sensors, muscle signals, and brain signals, with the goal of developing robots adapt to people rather than requiring people to adapt to robots. ^[35][36]

Applications include robot assistants for physical tasks, wearable technologies for navigation or communication, autonomous and assistive mobility systems, collaborative robots for manufacturing and healthcare, and systems for correcting robot behavior during shared tasks. ^[37][38]

Physical AI and efficient on-device intelligence

A second major thread of Rus's work concerns Physical AI: artificial intelligence that is embedded in machines and devices that sense, move, learn, adapt, and act in the physical world. In a 2024 TED talk, she described the convergence of AI and robotics as enabling AI to move beyond digital confines and engage dynamically with the real world, a direction she frames as Physical AI or physical intelligence. ^[39]

Physical AI places different demands on machine learning than many purely digital applications. A robot, vehicle, medical device, drone, or scientific instrument may need to operate locally, with low latency, limited energy, changing conditions, privacy constraints, and safety requirements. Rus has argued that measures of progress in AI beyond model size include efficiency, adaptability, inspectability, and how much useful intelligence a system can deliver per watt. ^[40][41][42]

One line of work in this area is liquid neural networks, developed with collaborators including Ramin Hasani, Mathias Lechner, Alexander Amini, and Radu Grosu. Inspired in part by compact biological nervous systems, liquid networks are continuous-time models whose internal dynamics can adapt as new inputs arrive. They have been applied to sequential decision-making and autonomous systems that must respond to changing physical environments. ^[43][44][45]

This research helped lead to Liquid AI, an MIT CSAIL spinout co-founded by Rus, Hasani, Lechner, and Amini. Liquid AI develops efficient general-purpose AI systems and liquid foundation models, with an emphasis on capable, aligned, trustworthy, and compute-efficient AI that can operate closer to devices and real-world applications. ^[46][47]

AI for scientific discovery

Rus has also worked on AI as a scientific instrument for observing and understanding complex natural systems. In this framing, AI does not replace scientific inquiry, but extends it by detecting structure in data that is too large, subtle, or multidimensional for unaided human perception. ^[48][49]

Rus is involved in Project CETI, an interdisciplinary scientific and conservation initiative that combines machine learning, robotics, biology, linguistics, cryptography, and ocean science to study sperm whale communication. Public descriptions of the project identify Rus as a principal investigator in machine learning and robotics, and The Audacious Project describes her as one of the experts involved in the project team led by David Gruber. ^[50][51]

Project CETI builds on a lineage of whale-song research associated with Roger Payne, whose recordings of humpback whales helped inspire modern whale conservation. CETI was formed in 2020 with catalyst funding from The Audacious Project and aims to use advanced machine learning and gentle robotic sensing to listen to and analyze sperm whale communication in behavioral and social context. ^[52][53][54]

In 2024, Project CETI researchers proposed a sperm whale phonetic alphabet based on machine-learning analysis of thousands of codas in social context. The result did not translate whale language, but it identified contextual and combinatorial structure in sperm whale communication and illustrated how AI can reveal patterns in nonhuman communication systems. ^[55][56]