Researchers led by Michael Maravgakis, with Stylianos Piperakis and Panos Trahanias, have built SBEE, a real-time open-source system that estimates a legged robot's position and the height of the ground together, using only its own body sensors, according to a paper on arXiv. When cameras cannot see well, such robots cannot observe their global position, and drift in their height estimate can upset their leg movements.

SBEE uses the ground itself as a fixed reference. Each stable footstep adds a height measurement to a grid map of the ground, and cells mapped earlier are fed back to correct the height estimate, much as loop closures do in mapping with cameras. Its filter uses a velocity measurement from the legs' kinematics, so its size does not grow with the number of legs.

The team tested it in simulation on four-legged and two-legged robots and on the ANYmal-D GrandTour dataset, covering 296 m of walking, where it cut the absolute trajectory error by 22.1% compared with the best of three leading estimators that rely on body sensing alone.

Source: arXiv Robotics (cs.RO)

Background

Michael Maravgakis

Robotics researcher

Robotics researcher.