Research in the Dynamics Robot Systems Group
The research we do spans a range of areas including motion planning state estimation, mapping, control and learning – typically focused on legged or dynamic robots.
Motion Planning: Walking robots typically have many more degrees of freedom (DOF) than typical fixed-base robots. This gives advantages of redundancy – making it possible to reach to a certain goal in many different ways. However it also brings complexity as the high number of DOFs (e.g. 12) and the requirement to remain balanced mean that typical planning approaches designed for low dimensional robots do not scale. In addition when walking there is the issue of making and breaking foot contact which makes the problem both discrete and continuous.
State Estimation and Mapping: In robotics a core problem is simultaneous localization and mapping (SLAM). We research estimation techniques focus towards walking robots. The challenge in this domain is that foot impacts and sudden motions of the robot cause many common methods to fail. In addition estimation and mapping on dynamic robots involves fusion of different sensor sources such as inertial, kinematic, LIDAR, vision, radar and force sensors. We have applied these methods to robots such as the Boston Dynamics Atlas, IIT’s HyQ and our current robot the Anymal.