Researchers at the Massachusetts Institute of Technology have developed a better way to control robots with soft physical bodies. This is quite the accomplishment as the physical structure of a robot can limit its capabilities.
While the field of soft robotics is relatively new, it is definitely an instrumental part of key areas in robotics. A robot with a soft body is an extraordinary feature specially for numerous tasks. These include surgeries or excavation that require accurate and delicate movement. Complex tasks like these require precision, and a soft robot might help in these situations.
Soft-bodied robots can move in a virtually infinite number of ways at any given moment. Optimizing them to perform specific tasks is a huge computational problem, but a new model from @MIT_CSAIL researchers can help. https://t.co/r2UmDUkeys pic.twitter.com/suPryG0aWz
— Massachusetts Institute of Technology (MIT) (@MIT) November 22, 2019
Developing soft body robots
However, soft robots are difficult to program. Computationally, creating a soft robot means that it can bend in an infinite number of ways to complete a task. It is imperative that the robot chooses the most effective movement since this also requires a lot of processing power. It was too difficult for a robot to have the computational power to calculate millions of dimensions and select an optimal pathway.
Thanks to MIT, they will reveal its soft robotics advancements during the ongoing Conference on Neural Information Processing Systems this week. Their model is able to learn a compact or “low-dimensional” state representation that is based on the physics of the robot, the environment, and other factors. The model is then able to co-optimize movement control as well as material design parameters. These are then aimed at specific tasks.
In the simulations that took place, the model enabled 2D and 3D soft robots to complete the target tasks. The tasks included moving different distances and reaching target spots. The model was able to do these faster and more accurately than other current methods. The researchers now want to use the model in real soft robots.
How they work
The robots use modern data compression methods to ensure that processing time and power requirements are relatively low. To accomplish this, all the input data gets compressed into smaller variables for the robot to consider. This ensures that the robot doesn’t take too long to process the inputs it is given.
The researchers validated their model on simulations of variations of 2D and 3D robots that are biped and quadruped. They found that, while robots using traditional methods can take up to 30,000 simulations to optimize these parameters, robots using their model took only about 400 simulations.
The robots are not ready for real-world use just yet, but it is no doubt a promising advancement. However, developing these robots for the real world would mean tackling issues with uncertainties and other factors that may come into play during specific target tasks. In any case, plans to develop a full program is in the pipeline, from simulation to fabrication, for soft robots.
YouTube: Learning-In-The-Loop Optimization: End-To-End Control And Co-Design of Soft Robots (NeurIPS 2019)