In a recent update from the Massachusetts Institute of Technology (MIT), researchers have developed a new AI method that grants robots the ability to use their entire bodies to manipulate objects, moving past the restricted use of just fingertips. This development, discussed in detail in a study published in the IEEE Transactions on Robotics on August 24, 2023, aims to enhance the efficiency and versatility of robotic systems in different sectors, such as manufacturing and space exploration.
Whole-Body Manipulation: A New Frontier in Robotics
Humans have mastered the art of utilizing their entire bodies to handle objects, effortlessly shifting from using fingertips to engaging arms and torso as needed. However, robots have historically struggled with such tasks, finding it challenging to compute the myriad potential contact events that occur during whole-body manipulation.
MIT researchers, led by H.J. Terry Suh and Tao Pang, have now simplified this complex process, known as contact-rich manipulation planning, using an AI technique called smoothing. This method condenses numerous contact events into a smaller set of decisions, allowing even basic algorithms to swiftly devise effective manipulation plans for robots.
Smoothing: The Key to Efficient Planning
The team discovered that smoothing, which averages out many insignificant intermediate decisions, enables robots to focus on crucial interactions and predict long-term behavior. This technique, combined with a specially designed algorithm, can generate complex plans in about a minute using a standard laptop, significantly reducing computation time.
This approach has proven to be as effective as reinforcement learning, a machine-learning technique where a robot learns to complete a task through trial and error, but operates much faster. “If you know a bit more about your problem, you can design more efficient algorithms,” notes Pang.
A Glimpse into the Future of Robotics
While still in its infancy, this method holds the potential to transform industries by allowing the use of smaller, mobile robots capable of manipulating objects with their entire bodies, thereby reducing energy consumption and costs. Moreover, this technique could be instrumental in space exploration missions, enabling robots to adapt swiftly to new environments using onboard computers.
Despite its effectiveness in slower manipulation tasks, the current model cannot handle highly dynamic motions, such as catching falling objects. The team aims to enhance the technique to address these dynamic challenges in the future.
Senior author Russ Tedrake emphasized the significance of this development, stating, “The same ideas that enable whole-body manipulation also work for planning with dexterous, human-like hands. Previously, most researchers said that reinforcement learning was the only approach that scaled to dexterous hands, but Terry and Tao showed that by taking this key idea of (randomized) smoothing from reinforcement learning, they can make more traditional planning methods work extremely well, too.”
This research, which heralds a new era in robotics, was funded in part by Amazon, MIT Lincoln Laboratory, the National Science Foundation, and the Ocado Group.
Reference
- Pang, T., Suh, H.J.T., Yang, L., Tedrake, R. (2023). Global Planning for Contact-Rich Manipulation via Local Smoothing of Quasi-Dynamic Contact Models. IEEE Transactions on Robotics. DOI