CCRL Explores Endless Learning & New Horizons in Robotics

Graduate Student Develops New Robot Training System

A graduate student of the Georgia Technological Institute, Tek Nirangan Kumar, has developed a fundamentally new robot training system called CCRL (Cascaded Compositual Residual Learning or Cascading Residual Training). This system allows a four-legged robot to perform more and more complex tasks without learning the basic movements again.

Kumar demonstrated the development of a heavy door opening transmission of energy, which is a truly outstanding achievement in the field of robotics.

The CCRL system functions like a “library” in which the robot can remember everything it learned when performing simple tasks. Each new skill is added to the library and serves as the foundation for acquiring more complex skills. For instance, once the robot learns how to turn while walking, it can further apply this skill for navigation.

According to Kumar, the CCRL system has uncovered new possibilities in interactive navigation studies, among several existing methods that enable robots to navigate in the real world. Other methods include point-to-point navigation and navigation towards specific objects.

Interactive navigation is considered the most challenging method as it requires interaction with obstacles along the way.

Kumar points out that the key to transitioning the robot from walking to object manipulation lies in the joints. The robot discovers various types of movements it can perform using these joints and successfully applies the gained experience in practice.

Currently, the CCRL system has mastered around ten skills. However, this number can significantly increase by using a more advanced hardware platform.

“The more skills added, the longer the training process, as the system needs to learn to apply them in different situations. But theoretically, new skills can be continuously incorporated as long as there is enough computational power,” explains Kumar.

According to him, CCRL will be beneficial for home assistants that need to navigate adeptly in cluttered environments, and it can also serve as a guide for visually impaired individuals by clearing paths and opening doors.

This development is considered a truly outstanding achievement that opens new horizons in the field of robotics.

/Reports, release notes, official announcements.