Motion Planning Based on Realistic Sensor Data for Six-Legged Robots

Abstract

Walking robots often use gait pattern to stride over the erth surface. In rough terrain this concept reaches its limitations, therefore reactive control mechanisms were introduced. In heavily unstructured terrain the robot is more reacting than actually moving. Motion planning based on gathered terrain information can help to reduce the necessary reflexes.

This article adresses the problem of real-time motion planning for walking robots. Therefore motion planning is regarded as optimization problem which is solved using a heuristical local search algorithm. It is described how we model the optimization problem how to solve it using random sampling.

Because realistic sensor data is less reliable the farther away the object is from the sensor we show how to break down the planning distance to small parts. Further more we show how to modify the algorithm to compose movements seamlessly so we can stop the actual movement and plan the next steps in case the robot encounters an obstacle the sensor has missed.

Experiments have shown the capabilities of the actual implementation but the tests have also schown the restrictions. The algorithm needs to be improved to find more good solutions in the given time, otherwise the locomotion gets interupted.