OpenAI releases robogym, a framework that provides simulated environments to train robots and enhance their capabilities. robogym uses other toolkits like OpenAI gym and MuJoCo physics simulator. While OpenAI gym offers an end-to-end suite of reinforcement learning tasks, Multi-Joint dynamics with Contact (MuJoCo) is a physics engine for robotics that facilitate research and development in a simulated environment.
Using robogym, you can not only visualize and interact with environments but also change the parameters of environments for diverse virtual settings. You can even teleoperate to manually manage a robot’s interaction using a keyboard, thereby giving varied capabilities to train robots to ensure it delivers superior performance.
As per Statista, global robotics market revenue will hit $100 billion, and by 2025 the market size will reach $210 billion. Machine learning, a major driver of the adoption of robots, is playing a significant role in the rapid adoption of robotics. Therefore, frameworks like OpenAI robogym will open up a wide range of opportunities for learners to quickly develop robots that can assist organizations as well as the general public in automating tasks.
Also Read: MIT Releases A Free Machine Learning Course
In 2019, OpenAI had made a breakthrough in robotics with its Automatic Domain Randomization (ADR) technique that allowed its robot — Dactyl — to train in different environments to solve Rubik’s cube. The simulator incrementally increased the complexity of the environment with randomization for the robot to train and enhance the dexterity of the robot.
It was one of the major developments in artificial intelligence as the technology shed light on machine learning’s general intelligence since the robot was able to perform in environments in which it was never trained for.
With OpenAI robogym, you can leverage the Dactyl environment that has a robotic arm with 20 actuated degrees of freedom to perform manipulation tasks. There are further sub-categories within this environment that can allow you to train robust robots with machine learning capabilities.