Novel Gaits for a Novel Novel Gaits for a Novel Crawling/Grasping - PowerPoint PPT Presentation

Novel Gaits for a Novel Novel Gaits for a Novel Crawling/Grasping Mechanism Crawling/Grasping Mechanism Richard Voyles Department of Computer Science and Engineering University of Minnesota AMAM 2000 University of Minnesota Department of Computer Science

Outline Outline � Motivation – DARPA Distributed Robotics Program – Rangers and Scouts � Design Details – Limb Mechanism – Integral Force/Torque Sensors � Locomotion Gaits – Swimming, Narrow, Wheel, Body-Roll � Gait Adaptation Strategy (future work) University of Minnesota Department of Computer Science

Scouts and Rangers Scouts and Rangers � DARPA Distributed Robotics Program � Hierarchical � Heterogeneous University of Minnesota Department of Computer Science

Ranger Macrobot Macrobot Ranger � Covers Long Distances � Remote Brain for Scout � Localization Capability � Carries and Launches 10 Scouts University of Minnesota Department of Computer Science

Scout Microbot Microbot Scout � Rolling/Hopping Locomotion � Compass, Tiltmeter, Communication � Camera / Microphone / Vibration / Gas Sniffer � Reconnaissance Surveillance Search/Rescue University of Minnesota Department of Computer Science

Scout Design Constraints Scout Design Constraints � Form Factor: hard constraint on body diameter (< 50mm) � Ruggedness: able to survive ballistic launching � Adaptability: encounters a variety of unknown conditions University of Minnesota Department of Computer Science

Scout Limitations Scout Limitations � No Ability to Manipulate (other than pushing) � Can’t Control Hop Height � Can’t Locomote when Headroom Only a Few Times Rubble Size University of Minnesota Department of Computer Science

Low Headroom Scenario Low Headroom Scenario � Terrorism � Warfare � Earthquake � Other Natural Disasters University of Minnesota Department of Computer Science

Alternate Design Goals Alternate Design Goals � Dual-Use Limbs � Same Form Factor � Same Ruggedness – Locomotion � Same Power – Manipulation Constraints � “Conserve Mechanism” � Greater Adaptability � Locomote by (low headroom) dragging the body � Dextrous Manipulation University of Minnesota Department of Computer Science

Outline Outline � Motivation – DARPA Distributed Robotics Program – Rangers and Scouts � Design Details – Limb Mechanism – Integral Force/Torque Sensors � Locomotion Gaits – Swimming, Narrow, Wheel, Body-Roll � Gait Adaptation Strategy (future work) University of Minnesota Department of Computer Science

TerminatorBot - - Alternate Scout Alternate Scout TerminatorBot � Two 3-DoF Arms that Stow Inside Body � Dual-Use Arms for both Locomotion and Manipulation � Four Locomotion Gait Classes: – “Swimming” Gaits (dry land) – Narrow Passage Gait (no wider than body) – “Bumpy Wheel” Rolling Gait – “Body-Roll” Dynamic Gait University of Minnesota Department of Computer Science

TerminatorBot Form Factor Form Factor TerminatorBot Stowed Configuration Deployed Configuration Hemispherical side for smooth manipulation Concave claw for traction/digging University of Minnesota Department of Computer Science

TerminatorBot Design Design TerminatorBot Drive Motors and Gearboxes (6 total) 1-Dof Elbow with Integral 2-DoF Shoulder Force/Torque Differential Sensor University of Minnesota Department of Computer Science

Elbow Force/Torque Sensors Elbow Force/Torque Sensors � Maltese Cross is isotropic but torque saturates � Top Figure has higher sensitivity to forces of interest, but torque saturates � Bottom figure “equalizes” torque � Use LVDTs to avoid cross- coupling (Khatib, et al) University of Minnesota Department of Computer Science

TerminatorBot Prototype Prototype TerminatorBot � 1st Prototype is 75 mm in Diameter (approx. 2x scale) University of Minnesota Department of Computer Science

Outline Outline � Motivation – DARPA Distributed Robotics Program – Rangers and Scouts � Design Details – Limb Mechanism – Integral Force/Torque Sensors � Locomotion Gaits – Swimming, Narrow, Wheel, Body-Roll � Gait Adaptation Strategy (future work) University of Minnesota Department of Computer Science

Swimming Gait Video Swimming Gait Video � “Expected” Gait � Variable Body Height � Adaptable to Terrain University of Minnesota Department of Computer Science

Narrow- -Passage Gait Video Passage Gait Video Narrow � New Mechanisms Often Suggest Novel Gaits (Yim, Xu, Pai) � Non-Controllable Body Height � Good Traction � Arms Stretched Out in Front of Body University of Minnesota Department of Computer Science

Bumpy Wheel Gait Video Bumpy Wheel Gait Video � Non-Controllable Body Height � 4 Coupled Motors � Forearms “Roll” Like the Spoke of a Wheel University of Minnesota Department of Computer Science

Body- -Roll Gait Roll Gait Body � Use Dynamics to Roll the Body � Requires Smooth Surface � Swing One Arm Across Body � Use Other Arm as Reaction Force � Tuck Swinging Arm to Roll � Un-implemented at this time University of Minnesota Department of Computer Science

One- -Armed Gaits Armed Gaits One � Fail-safe Operation – Emergency Homing Measures – Graceful Degradation of Mission Performance � Body-Roll Gaits University of Minnesota Department of Computer Science

Outline Outline � Motivation – DARPA Distributed Robotics Program – Rangers and Scouts � Design Details – Limb Mechanism – Integral Force/Torque Sensors � Locomotion Gaits – Swimming, Narrow, Wheel, Body-Roll � Gait Adaptation Strategy (future work) University of Minnesota Department of Computer Science

Locomotion Primitives and Skills Locomotion Primitives and Skills � Patterned after prior work in manipulation (Morrow, Voyles) � Gaits = Locomotion Skills – Gaits composed of collections of primitives – Gait cycles through primitive space University of Minnesota Department of Computer Science

Simple Primitives Simple Primitives � Open and Closed Loop – Joint-space trajectory segment – World-space trajectory segment – Torque command – Guarded move (sensor response) � Evaluation Metrics � Based on Port-Based Adaptable Agent Architecture (PB3A) University of Minnesota Department of Computer Science

Learn to Evaluate Gaits Learn to Evaluate Gaits � Metrics – Vertical Visual Servoing Error – Kinesthetic Sense of Torque – Visual Odometry � Eigenspace-Based Learning Approach (PBD) � Currently Applying Towards Conventional Mobile Robots � Morph Gaits from Closed- to Open-Loop University of Minnesota Department of Computer Science

Summary and Future Work Summary and Future Work � Robot Design � Novel Gaits � Topics Under Investigation – Gait Adaptation to Terrain • Learning Progress Metrics • Gait Morphing – Learning Manipulation Primitives • Primordial Adaptation (Eigenspace Method) • Programming by Demonstration University of Minnesota Department of Computer Science

Sponsors Sponsors � DARPA / MTO – Distributed Robotics Program – contract MDA972-98-C-0008 � Air Force Research Lab – Self-Adaptive Software Program – contract F30602-96-2-0240 University of Minnesota Department of Computer Science