DynoClimber: The Bipedal Dynamic Climber

research in collaboration with Prof. Jonathan Clark, Florida State University


Overview

Geckos, cockroaches, and hordes of other climbing animals are able to ascend vertical surfaces with staggering speed and agility that seems to match their capabilities on level ground. Research conducted at Berkeley’s Poly-PEDAL Lab by Dan Goldman examined the forces which geckos and cockroaches generate while climbing. It is this work that has informed and inspired the design of our climber.

DynoClimber was built to emulate the force patterns exhibited by climbing animals in the hopes of generating rapid, stable vertical locomotion. The robot is purpose built in its current incarnation to decouple the dynamics of climbing rapidly from the climber’s attachment to the wall; it climbs a carpeted vertical surface, employing aluminum claws to grasp the surface. This goal differs from the more utilitarian RiSE robot, which climbs a variety of different surfaces, albeit at a much slower pace.

Thus far, DynoClimber has achieved strikingly rapid dynamic climbing, achieving velocities of 66cm/s (1.5 bodylengths/second) up a vertical surface. Ongoing work by Goran Lynch in collaboration with Prof. Jonathan Clark includes the creation and testing of different control paradigms, as well as the empirical study of the robot’s stability properties.

 Sub-Projects

Self-exciting, work-directed control

DynoClimber is an ideal platform for the creation and testing of new legged control paradigms. Its actuated configuration space is remarkably small (T2), while the behavior it exhibits is dramatic and impressive (vertical locomotion at 1.5 bodylengths/second with substantial lateral oscillation of the robot’s mass center).

The two fundamental goals of a controller for our robot are:

Maintain attachment of at least one leg at all times

Maximize the mechanical power injected into the system

Extensions to Other Platforms

Kod*Lab students have spent substantial time tuning (or building algorithms to autonomously tune) feedforward gaits; RHex and similar robots owe much of their agility and speed to these well-tuned feedforward gaits. RHex and RiSE both employ a Central-Pattern Generator extensively in their behaviors.

The controller employed on DynoClimber has properties which are desirable in other legged control platforms. Specifically, myriad other legged robots might benefit substantially from a control architecture which permits a designer to specify work-targets.

External Links

Relevant Publications

2009

  • A Self-Exciting Controller for High-Speed Vertical RunningSubmitted Mar. 2009

2008

  • A Bio-inspired Dynamical Vertical Climbing RobotSubmitted Dec. 2008

2007

  • Design of a bio-inspired dynamical vertical climbing robotRobotics, Systems and Science, 2007 PDF Version