Research and Education Agenda for Howie Choset, Biorobotics Lab, Carnegie Mellon.

Intro / Snake Robots

Howie Choset has led a comprehensive research program in bio-inspired robotics and AI since he started as a faculty member at Carnegie Mellon in 1996. Perhaps the work for which Professor Choset is best known are his group’s snake robots. This work has garnered much media attention, such as being the first robot to appear on Late Night TV, as well as appearing on Good Morning America, back in 1998. CMU has asked Choset to speak on several occasions about the snake robots, including CMU’s first appearance at the World Economic Forum in 2011 and many appearances on the Hill. The notoriety is only the tip of the iceberg. Choset’s true contributions lie in the fundamental science his group pioneered, as well as, in the systems, his staff and students, designed, built, and deployed in real-world settings. Choset’s group runs the pipeline from basic research, to deployment and sometimes to commercialization, which in turn informs the basic research.

Geometric Mechanics

The core challenge Professor Choset’s group faces lies in coordinating the motion of all of the joints of the mechanism to produce purposeful motion. This is especially challenging for snake robots because they have many more joints, often called degrees of freedom, than conventional robots. To tackle this challenge, Choset’s group takes recourse to fundamental principles in a branch of mathematics called geometric mechanics. Geometric mechanics is essentially a combination of differential geometry, group theory and vector calculus. We certainly are not the first to investigate Geometric Mechanics, and it was brought to us from researchers at Caltech, where Professor Choset obtained his PhD. Our contribution to Geometric Mechanics can succinctly be stated as bringing a robotics kinematics perspective to the field, allowing us to advance the field so as to solve both low and high degree of freedom problems that generate motions for snake robots, and other systems, to locomote.

Professor Choset’s students, Drs. Elie Shammas (American University of Beirut), Ross Hatton (Oregon State University), Chaohui Gong (Bito Robotics) and Tony Dear (Columbia University) all have made or continue to make contributions to this field. It is worth highlighting Ross Hatton’s work which was able to address fundamental questions posed by a Nobel Laureate, and then answered by Professor Hatton in his PhD work. Dr. Hatton continues to advance the field of Geometric mechanics, and is considered to be the thought leader in this field. Professors Hatton and Choset are co-authoring a book on this subject to be used in their classes at their respective universities. Much of this book is Professor Hatton’s framework on Geometric Mechanics which makes the field highly accessible to new Engineering and Computer Science students. It will be the must-have book for robot locomotion, once it is published.

Bio-inspired Robotics / Robot-inspired Biology

Many researchers investigate biological systems in hopes of finding clues or motivations on how to build and program robotic systems that operate in challenging terrains. Choset’s group differs from these so-called biologically inspired robotics research groups in that his group seeks to use the geometric methods, described above, to model biological behavior and then once understood, try to replicate the behavior on a robot. However, something more provocative emerged: since Choset’s group took fundamental recourse to understand the mechanics of what makes the robot move, his group has made contributions to understanding biological locomotion; in other words, Choset’s group pursued a reverse illumination on using robots to understand biology. As such, his group, working in close collaboration with Professor Daniel Goldman at Georgia Tech, has advanced the fundamental math to model the motion of sand swimming lizards, sidewinders, and mudskipper fish. Baxi Chong, an MS student with Professor Choset, who continued with Professor Goldman for his PhD, also continues this work as a Professor at Penn State. Currently, Professor Choset is working with Professor Scott Kelly at the University of North Carolina to use the geometric methods to model how fish and other organisms communicate in swarms.

Search and Rescue

The concepts developed in Professor Choset’s group have also been implemented on real robot systems. This is important because experiments and deployments stress test the core assumptions, many times inadvertently, that were made when deriving new frameworks for robotics. Moreover, we all want to have societal impact with our work, and professor Choset’s group certainly has. Professor Choset’s group also deployed their locomoting snake robots in real search and rescue scenarios, including the immediate aftermath of the Mexico City Earthquake, where his research group, at the invitation of the Mexican Red Cross, cleared buildings with their robots. Virtually all of Professor Choset’s field deployments and theoretical grounding in Search and Rescue comes from Professor Robin Murphy, formerly of Texas A&M and now CMU. Her guidance in connecting theory and mechanism to reality has been priceless. Our experiences in breaking the snake robots in the field drove us to develop modular systems, described below.

Archeology

Professor Choset’s group also sent robots into caves, off the coast of the Red Sea and underneath the pyramids in Egypt to locate archeological artifacts; we found nothing. One of the challenges we faced in Egypt was enabling the snake robot to climb up sandy hills, a challenge we overcame when we started working with Professor Goldman’s group at Georgia Tech. This demonstrates how real-world deployment shines light onto problems otherwise overlooked.

Underwater Robotics

Professor Choset’s group built an underwater snake robot that has been deployed in swimming pools, the Allegheny River and a port in the Pacific Ocean. In addition, working with JPL, Choset’s team helped develop another underwater snake robot, to be deployed on one of the moons of Saturn. This moon has a very large underground ocean into which the robot would be sent, with the hopes of finding signs of life. While this program was indeed canceled, Choset’s team did deploy a robot in an ice skating rink in Pasadena and a glacier in Canada.

Medical Robotics

Professor Choset and his students developed a small surgical snake robot whose success led Professor Choset to co-found, along with his post-doc Dr. Alon Wolf (Technion) and medical collaborator Dr. Marco Zenati, MD (Harvard) a medical robotics company called Medrobotics. Medrobotics snake robot was the first medical robot to clear the FDA with the indication “robot.” We operated on over 1500 people in 4 continents and 17 countries. Among our largest investors were the Petersen Family, local to Pittsburgh and Western Pennsylvania.

Body-SLAM

Real-world deployment of the snake robots taught us a lot. The first lesson is simply answering the question: where am I? A considerable amount of effort has been dedicated to addressing the localization problem in robotics. Choset’s group differs from the others in that we develop localization algorithms (and sensors) for confined spaces, for applications specifically dedicated to medicine, search and rescue and manufacturing. Professor Choset’s students Drs. Stephen Tully (Activ Surgical) and Arun Srinivatsan (Apple) developed novel filtering methods that tolerate large initial uncertainty and yet converge to the correct estimate quite quickly. Dr Srivtastan’s work, in particular, incorporated the use of dual quaternions and Bingham filters that allowed us to prescribe a linear filter in rotational space, as opposed to linearizing on such spaces. Today, Dr. Srivatsan is one of the research leaders at Apple.

Medical AI

Ultrasound is another sensing modality that Professor Choset’s group, along with Professors Artur Dubrawski and John Galleotti at CMU, has been using to close the loop with medical diagnostics and therapy. Ultrasound is low cost, portable and emits no radiation, and yet is difficult to interpret and requires mechanical interaction between the patient’s anatomy and caregiver. Professor Choset’s group developed a novel needle insertion mechanism that uses both force feedback and ultrasound to guide the needle into a vessel. The guidance system is a machine learning model that identifies the vessel and directs the robot to aim the needle toward such vessel. Unfortunately, as with most machine learning approaches, this approach requires data, which is virtually non-existent because labeled ultrasound data does not exist in the wild. Instead, Professors Galleotti’s and Choset’s students developed a novel synthetic data generation technique that provides data of vessels being compressed at a large range of pressures, thereby allowing the model to be trained and the needle to be inserted. This approach has been validated on many live pig models at the University of Pittsburgh.

Modularity

The second lesson we learnt from deploying the snake robots lies with modularity. Already, Professor Choset’s thinking had been inspired by Professor Mark Yim at the University of Pennsylvania and Choset’s early snake designs were based on Yim’s. Modularity became important as a result of Professor Choset’s group cycle of: build, deploy, break, re-pair. Instead of re-building a new system, Professor Choset’s students Matt Tesch (Hebi Robotics) and David Rollinson (Hebi Robotics) decided to create modules, both in hardware and software, that enabled Choset’s team to just repair a module, and not the entire system. By developing and committing to interfaces among modules, Choset’s group was able to accelerate the development, repair and maintenance of novel robots because one need only design a module. This architecture led Choset, along with Rollsinson, Tesch, and two others Layton and Enner, to co-found Hebi Robotics to commercialize this technology. Hebi has customers and employees from all over the world, now based in Pittsburgh.

The work in modularity also runs the pipeline as well

Dr. Whitman (Boston Dynamics) along with MS students Mr. Jeff Hu developed novel machine learning models to address the problem of modular robot design. Their work addressed: given a set of modules, generate the optimal controller and arrangement of modules to execute a locomotion task. Naturally, we demonstrated the ideas of Dr. Whitman’s work on the Hebi modular robot system, as well as a new modular robot architecture, called Eigenbot, a project led by Li in Choset’s group.

Multi-Agent Planning

Just as snake and modular robots have many degrees of freedom, so do multi-agent systems and swarms, which also possess the challenge of coordinating all of the degrees of freedom to provide purposeful motion. Professor Choset’s students Drs. Glenn Wager and Richard Ren developed a novel multi-agent planning paradigm that, on average, beats the so-called curse of dimensionality, where two robots are twice as hard as one, three robots are four times more difficult, four robots are eight times, and so forth. Dr. Ren went on to address a large variety of multi-agent planners, including multi-objective multi-agent planning. Currently, he is writing a textbook on this topic, which will be the go-to text for all aspiring students in multi-agent planning. Recently, Mr. Anoop Bhat, Choset’s current PhD students, in cooperation with Professor Siva Ramithian at Texas A&M, have been developing all sorts of varieties of multi-agent traveling salesman problems, with the added complication that the targets to be visited by the agents move. This will have several applications in the Operations Research community.

Information-based Search

Professor Choset’s research group also investigates multi-agent search based on information that is either known previously or acquired at run-time. Choset’s graduate student, Ms. Ananya Rao, co-advised with Professor David Wettergreen, along with Choset’s post-doc, Dr. Bhaskar Vundurthy, are advancing the field of ergodic search, where agents use an a priori information map to guide the team of agents to balance the time searching for targets in regions based on the probability of finding a target in that region. This approach addresses the so-called exploration versus exploitation challenge commonly found in most optimization-based methods. Choset’s work differs from prior work in that it considers heterogeneous agents, and can search for multiple objectives at once. Another distinguishing feature of our work is that it uses classic geometric methods, used in Choset’s PhD thesis.

Manufacturing in Confined Spaces

Professor Choset’s group also developed large and strong snake robots for manufacturing in confined spaces with Boeing. Just as we developed mechanisms and algorithms for confined spaces, we also develop sensors with edge computing in confined spaces. The Boeing project initiated Professor Choset’s group’s work to develop such sensing capabilities. This project, led by Mr. Lu Li (Pipe Force / CMU) in Professor Choset’s group, uses conventional sensing modality, such as structured light, combined with novel custom developed electronics and software to enable point cloud data to be generated in tight spaces where conventional range sensors would not operate. Recently, Mr. Li, along with Ms. Tian (Pipe Force), an MS student in Choset’s group, co-founded a company called Pipe Force, AI, whose mission is to provide high resolution geo-synchronized digital twin maps of underground pipes. These maps will then be used to assess the state of such underground infrastructure allowing repairs to be planned proactively as opposed to reactively which is time consuming and expensive. Already, we have deployed our system in Texas and will begin mapping the underground storm pipes at CMU in the fall.

Decentralized Manufacturing / Space Robotics

As with our other projects, Choset’s students and staff built a group of multi-agent modular mobile bases for decentralized multi-robot assembly of aircraft wings for Boeing. This decentralized planning work is now being advanced to enable construction of large structures in space. One of the physical challenges for on-orbit assembly comes down to putting a peg in a hole. This requires careful coordination of the robot “feel” its way into a hole. To prescribe such algorithms, we developed a “holodeck” system that has four robot arms situated on rails. The combined motion of the robots on the rails mimics motion of objects in space on which we can test our controllers for peg-in-hole operations. This work is being done in cooperation with Northrup-Grummun.

Additive / Edge Manufacturing

Professor Choset’s research group has been collaborating with aerospace manufacturers such as Boeing, Northrop-Grumman and Raytheon to pioneer robotic and AI technology that enables additive and intelligent manufacturing. Their work, which has garnered several best paper awards at prestigious robotics and automation conferences such as ICRA and IROS, focuses on two key innovations. The first is edge sensor-based planning, which uses a custom-designed, close-range 3D and tactile sensor to acquire multimodal information, such as color, shape, force and contact, for in-situ inspection and monitoring. This allows for real-time motion adjustments and adaptive path replanning, ensuring precision and quality. The second innovation is the use of AI for intelligent in-process additive repair workflows. This approach utilizes edge sensor information to characterize material properties, enabling closed-loop process control, to reduce manufacturing defects and material waste. The group is currently working with industry partners to deploy these technologies on next-generation airplane and engine production lines, to promote a safer and eco-friendly future transportation.

Painting and Coating

Many applications require the deposition of spray paints or coatings. Most often, the goal is to achieve complete, uniform coverage of a precise thickness. This quality can be challenging to achieve even in simple planar surfaces, let alone complex curved surfaces or within confined spaces. This is important in the automotive industry where conventional approaches require three to five months to program a robot to “cover” a car. We have completed our work with Ford, and now Professor Choset’s group works with Boeing to program ink-jet applicators to create beautiful designs while painting, really ink-jetting, aircraft.

Abstraction

In many of the above described projects, a common theme emerges: reducing complicated problems to simpler ones so as analysis, design, and planning can take place in a tractable and practical manner. Starting with Choset’s PhD Thesis on deformation retractions with Voronoi Diagrams, his work reduced high-dimensional searchers to one-dimensional searchers. Likewise, the above-mentioned multi-agent planners search in one-dimensional spaces, only graduating to higher dimensional ones, as needed.

This abstraction theme permeates the machine learning work of Mr. Ben Freed, whose ultimate goal is that of forming simplified representations of complex problems that lend themselves to planning and decision-making. Consider the problem of driving to the store to buy a loaf of bread. This task requires the completion of many precise motor commands, completed in the correct sequence, making planning difficult. However, given a set of skills (e.g., a driving skill, a walking skill, and an object manipulation skill), the planning problem can be greatly simplified by planning over skills instead of raw motor commands. Mr. Freed, along with his co-advisor Jeff Schneider, is developing methods for automatically extracting such useful task representations. Here, we try to learn the low-level skills and the high-level policy to combine them, as a byproduct of the model learning procedure so that the skills are naturally aligned with the model of the dynamics and environment. The task of planning becomes far simpler if one can ignore the low-level details and therefore planning is abstracted and simplified.

Additionally, the machine learning work of graduate student Mr. Albert Xu tries to abstract away the specific kinematic details of a robot to capture the essence of its motion and enable the skills performed by one robot to be reproducible on other robots’ hardware. Conventionally, robot configurations are described by their joint angles and are specific to their particular robotic embodiment. Thus, it is often difficult to use one robot’s demonstration to teach a different robot how to perform the same task. Mr. Xu, along with his co-advisor Jeff Schneider propose to parameterize a robot’s configuration using a learned shape space that is shared between different robot embodiments to enable direct transfer of demonstration trajectories between different robots.

Education

One of the cornerstones of Professor Choset’s work is making the fundamental math, probability and algorithms easily accessible - so much so, that newcomers to these fields underestimate the contributions because they seem “so easy.” His presentations tend to be informative, educational, and often entertaining. This has carried over to Professor Choset’s classes, where he developed and continually taught undergraduates for nearly 20 years. Professor Choset created the Robotics Minor in 1998 and the Second Major in Robotics in 2007. According to the letters that supported Professor Choset’s nomination for the coveted Doherty Award, undergraduate education at Carnegie Mellon is a result of Professor Choset’s dedication and hard work toward undergraduate education. Choset’s dream aspiration is to package the educational experiences so that they can be used at Universities, other than Carnegie Mellon, with the hopes of impacting community college education.

Outreach

Finally, Professor Choset gives presentations to young people in K-12 schools, and on occasion to the elderly. In 2025, Choset hosted a refugee from Afghanistan who was part of the all-girl robotics team, called the Dreamers, which was shut down when the Taliban regained control. The girls and their families had to escape Afganastan. In 2023, Angel Studios produced a movie called Rule Breakers which was a dramatization of how this team formed. Two of Professor Choset’s students, Ananya Rao and Yizhu Gu, and one of his robots, Peggy, appeared in this movie. Both Ms. Rao and Gu now have IMb pages. This movie celebrates not only girls in STEM, but overcoming hardships to pursue one’s dreams, which in this case was robotics. In exchange for the students and robot, the producers let Professor Choset hold a screening in Pittsburgh where members of the CMU, Pitt, and CCAS communities were invited for a free preview of the movie. Professor Choset’s student from the Dreamers gave a powerful speech, and then Ms. Patti Rote spoke about the All-Girls team, Girls of Steel, that she, along with Professor George Kantor, started and mentored for many years. It was an incredible evening.