According to Interesting Engineering, researchers from the Massachusetts Institute of Technology (MIT) and Nvidia Research have developed a new algorithm cuTAMP to speed up the robot's action planning.
This technology allows robots to complete complex tasks with multiple steps in just a few seconds by using the parallel calculation capabilities of the graphics processing (GPU). Thanks to that, robots in factories or warehouses can process and package objects of different sizes and shapes more effectively.
According to MIT News: "accounting is very useful in the industrial context because time is really important at the moment. If the algorithm takes a few minutes to find a plan instead of a few seconds, businesses will spend a lot of money."
When packaging into boxes, robots must consider many factors such as how to hold objects, adjust, avoid collisions and follow packaging order and many other constraints. The more complex the job is, the more processing options are needed.
cuTAMP solves the above problem by using parallel computing through Nvidia's Cuda platform, combining two powerful techniques: Sampling and optimization. Instead of randomly taking samples, cuTAMP focuses on more potential solutions that are capable of meeting the constraints of tasks, thereby significantly improving the quality of options.
After that, cuTAMP assessed the ability to avoid collisions, meeting the constraints on movement and tasks of each solution. Auditing updates and filters options continuously until a feasible, high-quality plan is found.
According to the researchers, cuTAMP has successfully tested on the robotic arm at MIT and the Human-shaped robot at Nvidia. It does not require training data like a machine learning system and allows solving completely new problems without prior contact.
