DARPA researchers unveil robotic system performance improvements


As part of the Defense Advanced Research Projects Agency’s (DARPA) Machine Common Sense (MCS) program, researchers last week offered demonstrations of robotic system performance improvements, showcasing advances in intuitive physics, actors intentional and spatial navigation.

By comparing the progress to that of a child’s cognition, the researchers aim to create computer models and simulated training to prepare the robots for real-world scenarios. This preparation focuses on recognition and interaction with objects, agents and places. Their efforts made it possible to quickly adapt to changing terrain, learn to withstand a dynamic load, etc.

“These experiments are important milestones that bring us closer to building and fielding robust robotic systems with generalized motion capabilities,” said Dr. Howard Shrobe, MCS program manager in the Office of Innovation at the information from DARPA. “Prototype systems don’t need large suites of sensors to deal with unexpected situations that might arise in the real world.”

An experiment at the University of California, Berkeley created a rapid motor adaptation algorithm that allowed quadrupedal robots to adapt when coupled with proprioceptive feedback – the sense of self-movement and position from the body. In this experiment, the researchers noted that the robots successfully navigated a range of real and simulated terrain, which could prove useful for military units with load-carrying and sensing.

Speaking of loads, researchers at Oregon State also showed that a bipedal robot could learn to carry dynamic loads with only proprioceptive feedback. In a simulated-to-real learning environment, a robot named Cassie learned common-sense behaviors, adapting its stride to incorporate adjustments in its load, such as sloshing liquids or balancing weights. The robot walked for several moments on a treadmill with different types of load. Before training, the robot fell immediately.

The collaboration between researchers at the University of Utah and the MCS team at Oregon State University has also created an algorithm that allows robots equipped with multi-fingered hands to grab previously invisible objects – when they are trained – in simulation. After training, a robot was able to grasp with over 93% real-world success on new objects, compared to the 78% offered by existing passive learning approaches.

Other areas of research included attempts to create a commonsense repository of knowledge capable of enabling robots to learn by reading the web and to answer natural language and image-based questions about social phenomena. common meaning. A machine-created scalable knowledge base to augment diverse representation of the world and a variety of approaches to learning the commonsense structure of human behavior and physics from video were also part of ongoing research.