Design and implementation of an autonomous climbing robot
- This dissertation describes the design, implementation, and experimentation of an autonomous free-climbing robot, Capuchin. The objective of our project is to create a multi-limbed robot capable of climbing vertical terrain autonomously using techniques similar to those used by human free climbers. When a "free" climber climbs a steep crag or an artificial climbing wall, she uses nothing else but her hands and feet to make contact with terrain features such as holes, cracks, ledges or protrusions. Unlike "aid" climbing, which uses special equipments, tools, or engineered features, free climbing only relies on friction at the contacts between the climber and the terrain. In order to make a multi-limbed robot climb in a similar way, four fundamental challenges must be addressed: robot design, sensing, motion planning and motion control. Our work focuses on robot design (including sensors) and motion control. However, our robot, Capuchin, is an integrated system including a simplified sensing system and a pre-existing motion planner running off-line. A good robot design can increase the inherent ability of the robot to climb complex terrain. It may also lead to better performance and make other issues easier, such as motion planning and control. A four-limb structure was chosen after consideration of the robot's capability and complexity. Simulation was used during the design process to optimize performance, in particular to maximize the workspace reachable by the end-effectors (fingers). Sensors have been selected to allow the robot to both acquire information about the terrain and control its motion. Each finger is equipped with a camera. Vision feedback allows the robot controller to accurately dock the finger on a terrain feature at a location computed by the planner. It also allows modifying a planned trajectory in real-time, when the terrain differs slightly from the model that had been used by the planner or when other small errors occur (for instance, if the robot slips slightly at a contact). Each finger is additionally equipped with a force sensor that gives the magnitude and orientation of the reaction force at a contact. The four force sensors are used by the robot controller to maintain the robot in quasi-static equilibrium, by adjusting the robot posture and the joint torques when needed, so that the reaction forces at the contact point continuously within their Coulomb friction cones. A two-stage motion planner previously developed by Bretl and Hauser for free-climbing and other multi-limbed robots navigating on challenging and irregular terrain is used in this work. This planner decomposes a climbing motion into a sequence of moves, each performed with a fixed set of robot-terrain contacts (this set is called a "stance"). The transition at the end of each move consists of either breaking a contact or making a new one. The planner first computes a sequence of stances. Next it computes a trajectory for the move to be performed at each stance. If it fails to find a move at one step, it considers another sequence of stances. The core part of our research has been the design of the motion controller. The main problem we had to solve is a multi-contact force control problem. One of our most important findings has been the following: for quasi-static climbing, it is not necessary, even not desirable, to continuously control the forces exerted by the robot at the contact points. Instead, it is preferable to continuously monitor these forces and perform joint torque adjustments only when some reaction forces get too close to the boundaries of the friction cones or to their maximal magnitude. This strategy was not obvious when we started our research. In fact, we first developed a motion controller that continuously adjusted joint torques to keep measured reaction forces as close as possible to the terrain normals at the contact points. But computing these adjustments is rather time consuming. Moreover, this approach leads the robot to perform delicate adjustments frequently. As a result, robot motion was neither as smooth, nor as reliable as we would have liked. Instead, our new approach, which we call "lazy" force control, leads to a faster servo rate and much smoother motion. Our experiments show that on average adjustments only amount for a small percentage (less than 10%) of the total time spent climbing. They also demonstrate that Capuchin can reliably climb vertical artificial climbing walls autonomously and can handle small errors in the terrain model used by the planner.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Stanford University, Computer Science Department
|Salisbury, J. Kenneth
|Salisbury, J. Kenneth
|Statement of responsibility
|Submitted to the Department of Computer Science.
|Thesis (Ph.D.)--Stanford University, 2012.
- © 2012 by Ruixiang Zhang
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...