A new study suggests that it may be a long time before we can safely interact with legged robots in the real world.
Recently Study published in the IEEE/RSJ 2022 International Conference on Robots and Intelligent Systems (IROS) has described the problems of testing and safety characterization of legged robots. Led by a research team at The Ohio State University, the research focused on this type of machine, which uses mechanical limbs instead of wheels to move. The research findings reveal that current models of legged robots do not always behave as predictably in real-life scenarios, making it difficult to anticipate their success or failure in tasks involving movement.
Anti-Intuitive and Complex System
Bowen Weng is a PhD student in electrical and computer engineering at Ohio State.
“Our work reveals that these robotic systems are complex and, more importantly, anti-intuitive,” said Bowen. “This means you can’t rely on the robot’s ability to know how to react in certain situations, so completeness of the test becomes even more important.”
The scientific community is calling for universal safety testing regulations for mobile robots as more and more of them perform more sophisticated tasks. The integration of robots and artificial intelligence into our daily lives highlights the need for standardized safety measures. Legged robots, in particular, pose a significant safety risk, as they are often made of metal and can reach speeds of up to 20 mph. When operating in real-world environments with humans, these environmental uncertainties further emphasize the need for stringent safety regulations.
“Testing is really about assessing risk, and our goal is to investigate how much risk current robotics poses to the user or customer while in working condition,” said Weng.
Weng noted that while there are currently several safety specifications for the legged robot implementation, there is still no consensus on whether to test it in the field.
Developing a New Framework for Testing Legged Robots
This study is the first to develop a data-driven, scenario-based security testing framework for leafy robots.
“In the future, these robots may have the opportunity to coexist with humans, and are likely to be produced collaboratively by many international parties,” said Weng. “So having safety regulations and testing is very important for the success of this kind of product.”
The study leverages sample-based machine learning algorithms to determine how simulated robots might malfunction during real-world testing. That was partly influenced by Weng’s experience as a vehicle safety researcher at the Center for Transportation Research, a partner of the National Highway Traffic Safety Administration.
The team evaluated the set of conditions that ensure the robot’s stability when navigating a new environment, which is considered one of the important factors in determining overall safety performance. Using algorithms derived from previous robotics experiments, the team designed several scenarios for the robot simulation.
One trial focused on examining the robot’s ability to move while performing tasks with different gaits, such as walking backwards or stepping in place. In another experiment, the researchers tested the robot’s stability when pushed with enough force to change its direction.
The results showed that one robot failed to maintain balance in 3 out of 10 trials when asked to increase its walking speed. However, another robot was able to remain upright in 100 trials when pushed from its left side, but fell in 5 out of 10 trials when the same force was applied to its right side.
While it may take time, the research framework has the potential to support commercial deployment of legged robots and provide a safety benchmark for robots with different structures and properties. Weng mentioned that it would take time before the framework could be implemented.
“We believe this data-driven approach will help create an impartial and more efficient way to carry out robotic observations in test environmental conditions,” said Weng. “What we’re working on is not immediate, but for future researchers.”