The new general-purpose optimizer can speed up the design of walking robots, self-driving cars, and other autonomous systems.
Courtesy of MIT News Office
Autonomous robots have come a long way since the annoying Roomba. In recent years, artificial intelligence systems have been introduced in self-driving cars, last mile food delivery, restaurant services, patient screening, hospital cleaning, food preparation, building security, and warehouse packaging.
Each of these robotic systems is a product of an ad hoc design process that is specific to that particular system. When designing an autonomous robot, engineers need to perform a myriad of trial and error simulations, often instinctively. These simulations are tailored to the components and tasks of a particular robot to tune and optimize the performance of that particular robot. In some respects, the design of today’s autonomous robots is like baking a cake from scratch, without recipes or prepared mixes to ensure successful results.
Today, MIT engineers have developed common design tools for robotics to use as a sort of automated recipe for success. The team has devised an optimization code that can be applied to simulations of virtually any autonomous robot system and can be used to automatically identify how and where to fine-tune the system to improve robot performance. ..
The team has shown that this tool can quickly improve the performance of two very different autonomous systems. One is a robot navigating the path between two obstacles, and the other is a pair of robots working together to move a heavy box.
Researchers hope that the new general-purpose optimizer will help speed up the development of a wide range of autonomous systems, from walking robots and self-driving cars to teams of soft and dexterous robots and collaborative robots.
The team, consisting of MIT graduate student Charles Dawson and MIT’s assistant professor of aerospace engineering, Chu Chu Fan, will present the results at the annual Robotics: Science and Systems conference in New York later this month.
Inverted design
After observing the wealth of automated design tools available in other engineering disciplines, Dawson and Fan realized the need for common optimization tools.
“If a mechanical engineer wants to design a wind turbine, he can use 3D CAD tools to design the structure and finite element analysis tools to see if it can withstand a particular load,” Dawson said. say. “But these computer-aided design tools for autonomous systems are lacking.”
Robotics typically optimize an autonomous system by first developing a simulation of the system and its many interacting subsystems, such as planning, control, perception, and hardware components. Next, you need to adjust the specific parameters of each component and run the simulation forward to see how the system runs in that scenario.
Only after a lot of trial and error and many scenarios can a robotic engineer identify the optimal combination of ingredients to produce the desired performance. What Dawson and his fans tried to bother with was a boring, over-tuned, time-consuming process.
Instead of saying, “What about performance when you think about design?” By reversing this, Dawson explains, “Given the performance you want to see, what’s the design to get there?”
Researchers have developed an optimization framework, computer code. It can automatically find the tweaks that can be made to existing autonomous systems and achieve the desired results.
The heart of the code is based on automatic differentiation, or “autodiff”. It was the first programming tool developed within the machine learning community and used to train neural networks. Autodiff is a quick and efficient way to “evaluate derivatives” of the sensitivity of computer programs to changes in parameters. Dawson and Fan have developed a general-purpose optimization tool for autonomous robot systems based on recent advances in autodiff programming.
“Our method automatically teaches us how to take small steps from the initial design to a design that achieves our goals,” says Dawson. “We use autodiff to essentially delve into the code that defines the simulator and understand how to do this inversion automatically.”
Build a better robot
The team tested the new tools on two separate autonomous robot systems and showed that the tools quickly improved the performance of each system in laboratory experiments compared to traditional optimization methods. ..
The first system consisted of a wheeled robot whose mission was to plan a route between two obstacles based on signals received from two beasons located at different locations. The team sought to find the optimal placement of the beacon to create a clear path between obstacles.
They found that the new optimizer quickly re-simulated the robot and identified the optimal placement of the beacon within 5 minutes compared to the 15 minutes of the traditional method.
The second system is more complex, with two wheeled robots working together to push the box towards the target position. The simulation of this system included more subsystems and parameters. Nonetheless, the team’s tools efficiently identified the steps the robot needed to reach its goals with an optimization process that was 20 times faster than the traditional approach.
“If you have a lot of parameters to optimize for your system, our tools will perform even better and save you a lot of time exponentially,” says Fan. “This is basically a choice of combinations. The more parameters you have, the more choices you have. Our approach is to reduce it all at once.”
The team plans to make the popular optimizer available for download, further refine the code, and apply it to more complex systems, such as robots designed to interact with and work with humans. doing.
“Our goal is to enable people to build better robots,” says Dawson. “We don’t have to start from scratch because we are providing new building blocks to optimize the system.”
This study was partially supported by the Defense Science and Technology Agency of Singapore and the MIT-IBM Watson AI Lab.
Check out our brand new E-BikeGuide.. If you are interested in electric bikes, this is the perfect place to start your e-mobility journey!
Thank you for CleanTechnica’s originality and cleantech news coverage? Consider becoming a member, supporter, technician, ambassador, or Patreon patron of CleanTechnica.
