First “Frontiers in Robotics” Conference Highlights Columbia Research and Innovation

Joel Stein, Simon Baruch Professor and chair of the Department of Rehabilitations and Regenerative Medicine at Columbia University College of Physicians and Surgeons, followed Dean Boyce and introduced the range of options for mechanized physical therapy. Wearable powered braces to compensate for paralysis, telepresence robots to aid the socially isolated, and “robotic gyms” were some of the robotic solutions that can ultimately offset the rising cost of having people fulfill these roles.   

Taking a deeper look into medical rehabilitative robotics, Sunil Agrawal, professor of mechanical engineering and of rehabilitative and regenerative medicine in the mechanical engineering department, described his use of robotics as “like training wheels,” allowing people to learn through repeated training and to remove the robotics aids when they become proficient. A visit to his lab provided demonstrations of his tethered pelvic trainer (T-Pad) for patients with Parkinson’s disease, an exoskeleton technology to help stroke patients strengthen their weaker sides, and C-Alex, a lightweight training exoskeleton that can be customized to different groups to retrain gait.

Focusing on “versatile manipulation,” which he called the “unsolved problem in robotics,” Matei Ciocarlie, assistant professor of mechanical engineering, had the audience reflect on their hands, which can sense shape, size, temperature, vibration, texture, and proprioception, as well as how much force is applied, in order to better understand the extent of this challenge. Using simple components—light-emitting diodes and photo receivers—mounted on an instrumented stage, he is developing an accurate, low-cost solution for artificial tactile sensing that can be applied for robotic manipulation in domains ranging from manufacturing and logistics to dangerous areas such as disaster relief. For robot manipulation in orbit, where space is at a premium, he introduced a robotic hand optimized to perform many tasks on board the International Space Station using a single motor. Overall, Ciocarlie demonstrated a number of current projects in his ROAM lab, as well as the theoretical work that underpins them.

Chad DeChant, a PhD student in computer science; Philippe Wyder, a master’s student in mechanical engineering; Jonathan Blutinger, a PhD student in the Creative Machines Lab; and Aslan Miriyev, a postdoc in soft robotics,  presented work from the lab of Mechanical Engineering Professor Hod Lipson, discussing and later demonstrating applications such as using low-cost drones and deep learning to fight crop disease, 3D food printing and laser cooking, and soft actuators to enable soft robotics.

Chad DeChant outlined the problem of northern leaf blight, which destroys 13 percent of corn crops in the U.S. and Canada each year, a loss of around $2 billion. Deploying drones to photograph corn fields and then analyzing those images using deep learning to pinpoint the disease location is a potential solution. Using heat maps of crops and images showing disease lesions, the system can be trained to 97 percent accuracy in identifying the affected crops, in some cases more accurate than people given the same task. One of the main hurdles is a regulatory one, the current FAA regulation that prohibits commercial drone operations out of line of sight. The FAA is currently working on solutions to relax this rule.    

Shifting from crops to food, Jonathan Blutinger presented 3D printing of food, which would allow a combination of ingredients to be personalized for one’s own nutritional needs. Another use of 3D printing was outlined by Aslan Miriyev, who talked about soft actuators for soft robots that could be 3D printed from an inexpensive combination of ethanol and an elastomer matrix, resulting in an artificial muscle operated by a soft-actuated motor. This new approach opens up many applications, from handling fragile objects—even eggs—to working with people. 

Peter Allen, professor of computer science and a member of the Data Science Institute, offered an optimistic outlook for robotics in the home, the hospital, and the cloud. Better hardware, better sensors, and better reasoning with better software are all helping to bring down the cost of robots. Big data, parallel grid computing, collective learning, crowdsourcing, and vast data sets are contributing to advancing robotic capabilities. Allen’s video of a laundry-folding robot gave a taste of the complexity of predictive simulation and led to a discussion of grasping. New machine learning techniques allow robots to grasp an object by pre-computing grasps on thousands of objects—the Columbia grasp database shows 300,000 grasps of 8,000 objects—to come up with the most probable grasp for an unseen object. Machine learning can also be used to infer the entire geometry of an object from a single view, allowing robots to grasp objects amid clutter and occlusion.  In the hospital, Allen showed a number of new surgical robots that can perform minimally invasive surgery and perform automated surgical tool tracking to assist a surgeon.

Concluding the presentations, Dennis Fowler, medical director of the Columbia Biomedical Accelerator, biomedical engineering department and CEO of Platform Imaging LLC, used his own experiences to show the challenges of commercializing new surgical robotic technology. He outlined the risks—market, competitive, technical, financial, management, legal, regulatory—as well as the importance of putting together a team that is credible to investors. He also noted that de-risking several categories of risks increases the probability of raising sufficient capital to achieve success. 

Lab visits followed these presentations, with demonstrations of the technologies discussed.