Robots that need to use their arms to make their way across treacherous terrain are being equipped with an algorithm that found successful paths three times as often as standard algorithms, while needing much less processing time. The new algorithm speeds up path planning for robots that use arm-like appendages to maintain balance on treacherous terrain such as disaster areas or construction sites.
The research enables robots to determine how difficult the terrain is before calculating a successful path forward, which might include bracing on the wall with one or two hands while taking the next step forward. Machine learning was used to train the robot on different ways it can place its hands and feet to maintain balance and make progress. Then, when placed in a new, complex environment, the robot can use what it learned to determine how traversable a path is, allowing it to find a path to the goal much faster.
However, even when using this traversability estimate, it is still time-consuming to plan a long path using traditional planning algorithms. The researchers used a “divide and conquer” approach, splitting a path into tough-to-traverse sections, where they can apply their learning-based method, and easier-to-traverse sections, where a simpler path planning method works better. To do this, they need a geometric model of the entire environment, which could be achieved in practice with a flying drone that scouts ahead of the robot.
In a virtual experiment with a humanoid robot in a corridor of rubble, the team’s method outperformed previous methods in success and total time to plan — both of which are important when quick action is needed in disaster scenarios. Over 50 trials, their method reached the goal 84% of the time compared to 26% for the basic path planner and took just over two minutes to plan compared to over three minutes for the basic path planner.
The researchers also showcased the method’s ability to work on a real-world, mobile manipulator — a wheeled robot with a torso and two arms. With the base of the robot placed on a steep ramp, it had to use its “hands” to brace itself on an uneven surface as it moved. The robot utilized the team’s method to plan a path in just over a tenth of a second compared to over 3.5 seconds with the basic path planner.
In future work, the team hopes to incorporate dynamically stable motion, similar to the natural movement of humans and animals, which would free the robot from having to be constantly in balance and could improve its speed of movement.
For more information, contact Dan Newman at