Title: “Localization and terrain mapping on StarlETH.”
Speaker: Péter Fankhauser
The success of walking robots in real world applications depends on how well we are able to exploit their (potential) advantage in mobility. In this presentation, we are giving an insight into the approaches we are taking with the quadrupedal robots StarlETH and ANYmal for precise localization and terrain mapping to autonomously walk in challenging environments. We show results from laboratory experiments as well as from the application in a real-world scenario during the ARGOS challenge.
Similar to other robots like Boston Dynamic’s ATLAS, we employ a rotating Hokuyo laser sensor for environment perception. First, the point cloud acquired in one full turn, which takes about 2 seconds, are dewarped. To this end, we use the robot-internal state estimation, which is a fast-updated (400Hz) filtered state from IMU readings as well as kinematic and ground contact measurements. Second, scans are matched using an ICP framework to establish a laser odometry as well as to create a continuously growing point cloud of the environment.
Since foothold selection often depends on very detailed geometries of the ground, our quadrupeds are additionally equipped with a downward looking ToF camera providing a pattern of depth measurements at frequencies up to 160Hz. By stochastically fusing this 3D information with the local state estimation of the robot, we can create a robot-centric elevation map including uncertainty.
Marco Hutter is Deputy Director of the Autonomous Systems Lab at ETH Zurich. After studying mechanical engineering, he conducted his doctoral degree with the focus on design, actuation, and control of legged systems at ETH. During his time at ASL, Marco realized electrically actuated, torque controllable quadrupedal robots that are capable of different gaits ranging from robust climbing to fast and energy efficient running. Recently, his group started working towards completely autonomous navigation in challenging environments which involves environment perception and motion planning.
Péter Fankhauser is a PhD student at the Autonomous Systems Lab (ASL) at ETH Zurich since 2013. He received his MSc degree in Robotics, Systems and Control from ETH Zurich in 2012. His research interests are in mapping and motion planning for legged robots.