Robots are increasingly being used for various applications including search and rescue operations and combat situations. Unfortunately, during the performance of maneuvers, a robot may fall or tip over, preventing it from moving normally. Controlling the robot to successfully right itself can be a difficult and time-consuming task for operators — a major problem in situations that may be both time-sensitive and dangerous; for example, those controlling the robot must determine how to re-orient the robot to a desired position, if it is possible.
There have been several approaches employed for robot self-righting that can be categorized into four main groups: passive approaches, specific mechanisms, overturned drivability, and dynamic approaches. Passive approaches do not make an effort to actively self-right, relying on the shape of the robot and its center of mass location to allow for easy righting or to inhibit flipping. Some robots rely on specific mechanisms for self-righting, such as a flipper.
Another category of robots allows for upside-down operation, attempting to limit the need for self-righting. And still other robots take a dynamic approach, focusing on the release of stored mechanical energy in an attempt to right itself, such as by leveraging spring legs or generating rolling momentum.
An improved robot self-righting methodology has been developed that could be applied to any generic robot. A computational methodology is executed to analyze various orientations of the robot and joint configurations to determine those that are stable and those that induce instability (i.e., tip-over) events. The results from this analysis may be organized into a graphical network model. This methodology allows robots to autonomously determine how to right themselves, to provide designers with a tool to assess whether their robots are able to self-right, and to determine the qualities that make robots more capable to self-right.
The calculations may be performed prior to fielding the robot, creating a static map of the conformation space that can be stored using a computer-readable storage medium. The sensory data from the robot is then used to localize the current state on the map, and a path plan need only be generated to move from the current state to the nearest pre-computed optimal path to implement the above described method.