HAPTIC CONTROL AND OPERATOR-GUIDED GAIT COORDINATION OF A PNEUMATIC HEXAPEDAL RESCUE ROBOT A Master’s Thesis Presentation By Brian A. Guerriero Georgia Institute of Technology George W. Woodruff School of ME Intelligent Machine Dynamics Lab NSF Center for Compact and Efficient Fluid Power Dr. Wayne Book CCEFP TB4 Brian Guerriero Introduction NSF CCEFP Paving the way in improving the compactness, efficiency, and effectiveness of fluid power 7 member universities 3 thrusts 4 testbeds CCEFP TB4 Brian Guerriero Introduction Testbed 4: Compact Rescue Crawler Develop testbed for man-machine multimodal interface research Research bilateral teleoperation and coordinated pneumatic control Research methods of enabling a single operator to control an 18 DoF mobile robot Use PHANToM haptic devices to wield control over two robot legs CCEFP TB4 Brian Guerriero Introduction CCEFP Collaborator Roles Vanderbilt University Develop chemofluidic monopropellant fuel source and components Develop high-level automatic gait coordination NCAT Evaluate human factors issues regarding operator interface Evaluate optimum methods for feeding large amounts of data effectively to operator CCEFP TB4 Brian Guerriero Acknowledgements Dr. Wayne Book Dr. Harvey Lipkin Dr. Chris Paredis JD Huggins Others: Dr. Matt Kontz IMDL Labmates Friends & Colleagues Dr. Haihong Zhu CCEFP TB4 Brian Guerriero Acknowledgements Industry Support and Sponsors CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero Background Research Pneumatic Servo Control Wang, Pu, Moore: acceleration feedback instead of pressure Chillari, Guccione, Muscato: Survey of pneumatic control schemes Differential pressure gain scheduling Fuzzy, Neuro-Fuzzy, Sliding mode Guvenc: Discrete time model regulation with model inversion Korondi and Gyeviki: robust sliding mode control CCEFP TB4 Brian Guerriero Background Research Chemofluidic Monopropellant Research Goldfarb, Barth, Fite, Mitchell, Shields, Gogola, Wehrmeyer: Control, characterization and implementation techniques Al-Dakkan, Goldfarb, Barth: Energy saving techniques reusing high-pressure exhaust gasses CCEFP TB4 Brian Guerriero Background Research High Level Gait Coordination Cruse: Stick insect cauausius morosus gait analysis, developed WALKNET Wait and Goldfarb: Further WALKNET development, application to a legged robot and simulations Torige, Noguchi, Ishizawa: Centipede style gaits moving feet in waves based on previous foot positions CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero CRC V1.0 Developed from Vanderbilt design 7/8” Airpel/Sentrinsic Actuators 3 DoF, Good Range of Motion CCEFP TB4 Brian Guerriero CRC V1.0 Dec. 06 – Apr. 07 Mounted to table Simple PID Control CCEFP TB4 Brian Guerriero CRC V1.0 Problems and Issues No-stiction cylinders proved difficult to control, 100 psi MAX Weak shoulder joint design Mechanical interferences CCEFP TB4 Brian Guerriero CRC V1.0 V1.0 In Action CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero CRC V2.0 4-07 - Present Complete and thorough two-legged redesign Designed for 300 psi actuation New prototype Sentrinsic cylinders CCEFP TB4 Brian Guerriero CRC V2.0 Design Benefits Shoulder Joints Clevis system eliminates slop and wear CCEFP TB4 Brian Guerriero CRC V2.0 Design Benefits Larger Actuators 1.5” pneumatic cylinders: 530 lbf at 300psi operating pressure Valves mounted on or as close as possible to cylinders CCEFP TB4 Brian Guerriero CRC V2.0 Design Challenges Range of Motion Decreased due to larger cylinders Prevent mechanical interferences Safety Robots Hurt! Integration Sensors, valves and actuators packaged together CCEFP TB4 Brian Guerriero CRC V2.0 Fabrication Aluminum Leg Profiles Waterjet cut at GTRI and finished at ME shop CCEFP TB4 Brian Guerriero CRC V2.0 Fabrication Senrinsic Cylinders Designed and built by Sentinsic at GT Custom rod ends and base clevises NFPA tie-rod design and fiberwound barrel construction Months of development, fabrication, debugging and revisions CCEFP TB4 Brian Guerriero CRC V2.0 Fabrication Senrinsic Cylinders 0-10V position output Integrated pressure sensors CCEFP TB4 Brian Guerriero CRC V2.0 Sensors Position CCRS integrated into cylinders Pressure Measurement Specialties 250 psi MEMS sensors CCEFP TB4 Brian Guerriero CRC V2.0 Sensors Pressure Tested for linearity Custom housings integrated into ends of cylinders CCEFP TB4 Brian Guerriero CRC V2.0 Sensors Sensor Integration Op-amp board developed for 12x pressure sensors Custom PCB routes all power, sensors and valve commands CCEFP TB4 Brian Guerriero CRC V2.0 System Integration Onboard Computing PC-104+ stack runs real-time control via xPC Target 802.11n wireless data transfer 32 16-bit Analog inputs 16 12-bit Analog outputs CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero Control Transformations PHANToM/operator Cartesian space Joint Space (θ1, θ2, θ3) Cylinder Space CCEFP TB4 Brian Guerriero Control Stroke Control Cylinder stroke length command converted into 0-10V command Festo Proportional Valves control flow into each cylinder CCEFP TB4 Brian Guerriero Control Goals Stability Pneumatic systems are high-order and traditionally difficult to control Tracking performance Each cylinder under highly varying loading conditions Target: 10% Robust to disturbances Noise and debris impacts CCEFP TB4 Brian Guerriero Control Original PD Control Control effort based on position error only Stable, worked well in original configuration yPD k p e kd e Vvalve yPD 5 CCEFP TB4 Brian Guerriero Control Two-Legged PD Control (Mounted) CCEFP TB4 Brian Guerriero Control Critical Flaw When weight applied to legs, control effort not high enough Large position errors Crawler could not actually crawl CCEFP TB4 Brian Guerriero Control Cylinder L1 Tracking Response 1 0.5 0 90 95 100 105 110 105 110 105 110 Stroke Length (in.) Cylinder L2 1 0.5 0 90 95 100 Cylinder L3 1 0.5 CCEFP TB4 Brian Guerriero 0 90 95 100 Time (s) Control Two-Legged PD Control (Struggling) CCEFP TB4 Brian Guerriero Control Improvements Addition of velocity feed-forward command Velocity damping term Differential pressure gain scheduler p kdp1 : p 0, e 0 yPD k p xref xact kd xact kvff xref CCEFP TB4 Brian Guerriero p k : p 0, e 0 dp 2 ydp p kdp 3 : p 0, e 0 p kdp 4 : p 0, e 0 Vvalve yPD ydp 5 Control Results Supplementary force control improved tracking Crawler developed a ‘hopping’ syndrome, decreasing stability Vvalve yPD ydp 5 CCEFP TB4 Brian Guerriero Control Hopping syndrome CCEFP TB4 Brian Guerriero Control Results Hopping caused by instantaneous gain change from position error sign change p kdp1 : p 0, e 0 p k : p 0, e 0 dp 2 ydp p kdp 3 : p 0, e 0 p kdp 4 : p 0, e 0 Vvalve yPD ydp 5 CCEFP TB4 Brian Guerriero Control Cylinder L1 1 0.5 Stroke Length (in.) 0 60 65 70 Cylinder L2 75 80 65 70 Cylinder L3 75 80 75 80 1 0.5 0 60 xref x actual 1 0.5 0 60 CCEFP TB4 Brian Guerriero 65 70 Time (s) Control Solution Scale force supplement by position error and differential force p kdp1 : p 0, e 0 p k : p 0, e 0 dp 2 ydp p kdp 3 : p 0, e 0 p kdp 4 : p 0, e 0 Vvalve yPD ydp 5 CCEFP TB4 Brian Guerriero F kdfe1 e ke : F 0, e 0 F k e k : F 0, e 0 dfe 2 e ydfe F kdfe3 e ke : F 0, e 0 F kdfe 4 e ke : F 0, e 0 Vvalve yPD ydfe 5 Cylinder L1 Control 1 xref xact 0.5 Stroke Length (in.) 0 45 50 55 60 65 60 65 60 65 Cylinder L2 1 0.5 0 45 50 55 Cylinder L3 1 0.5 0 45 CCEFP TB4 Brian Guerriero 50 55 Time (s) Control Improved force-based position control CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero User Interface Operator Workstation Reconfigurable task-space Initial Augmented Reality (AR) setup CCEFP TB4 Brian Guerriero User Interface Operator Tasks Feel environment and obstacles PHANToM haptic devices See and hear environment Head-mounted display PTZ camera onboard robot Ancillary functions Tactile switches on PHANToMs Voice recognition CCEFP TB4 Brian Guerriero User Interface Haptic Interface PHANToM force output Directional Proportional to position error Haptic Force, Y-axis, Full controller Spring force 4 3 Force (y-axis) (N) 2 1 0 -1 -2 -3 CCEFP TB4 Brian Guerriero -4 80 85 90 95 Time (s) 100 105 110 User Interface Immersive Environment Head-mounted display of feeds operator robot’s-eye-view Motion tracker translates head movements into camera movement CCEFP TB4 Brian Guerriero User Interface Head-Camera Interface CCEFP TB4 Brian Guerriero User Interface Ancillary Functions Operator communications with high-level gait controller Voice and tactile methods Visual robot status feedback Fuel Leg positions Noise alerts CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero Guided-Gait Coordination High Level Control Operator must wield control over 18 degrees of freedom WALKNET coordination ideal for smooth flat terrain and simple commands WALKNET coordination not sufficient for maneuvering through debris CCEFP TB4 Brian Guerriero Guided-Gait Coordination Tiers of CRC Control 1. WALKNET Gait Coordination Simple operator commands i.e. ‘Forward’ or ‘Left’ 2. Guided-Gait Coordination Operator haptically places front legs, rear pairs follow 3. Complete Control Operator haptically controls any leg (Extreme maneuvering) CCEFP TB4 Brian Guerriero Guided-Gait Coordination Guided-Gait Outline L1 R1 L2 R2 R3 L3 CCEFP TB4 Brian Guerriero Guided-Gait Coordination Trajectory Recording Record foot trajectories made haptically Smooth trajectory with a spline PHANToM x 200 300 0 0 2 4 6 8 10 PHANToM y 12 8 10 PHANToM z 12 14 16 18 200 300 y (mm) PHANToM input (mm) -200 Raw Trajectory Splined Trajectory 200 100 0 100 0 2 4 6 14 16 18 50 100 0 0 -150 -50 -100 0 2 4 6 8 10 Time (s) CCEFP TB4 Brian Guerriero 12 14 16 0 -50 50 18 x (mm) -100 150 -200 z (mm) Guided-Gait Coordination Stepping Stones Each trajectory Ti is a map between two known safe points Coordinate transforms relate robot position to inertial reference frame W pr updated each cycle (distance from origin) CCEFP TB4 Brian Guerriero Guided-Gait Coordination Successive Leg Pairs Recorded trajectories played through rear legs Master list of stepping stones and trajectories CCEFP TB4 Brian Guerriero Guided-Gait Coordination Conditions and Goals Maintain forward progress “Move rear legs to the most anterior reachable stepping stone” “Advance body until one leg reaches its posterior extreme point” Operator must tell coordinator when to move a set of legs Operator gives cue to the controller to begin body advancement CCEFP TB4 Brian Guerriero Guided-Gait Coordination Body Advancement Body advances by moving all six feet rearward at the same rates Advancement stops when one leg is at its posterior extreme position CCEFP TB4 Brian Guerriero Guided-Gait Coordination CCEFP TB4 Brian Guerriero Guided-Gait Coordination CCEFP TB4 Brian Guerriero Presentation Outline Background Research Pneumatic Control High Level Gait Coordination CRC V1.0 CRC V2.0 Design Sensors System Configuration Control Classical Methods Revised Force-based Position Controller User Interface Haptic Feedback Operator Workstation Guided-Gait Coordination Conclusions & Next Steps CCEFP TB4 Brian Guerriero Conclusions & Next Steps Robot Construction Two legged robot design is robust, easy to maintain, and reliable Future Work Stiffen spine Package computers & PSUs Integrate VU H2O2 technology CCEFP TB4 Brian Guerriero Conclusions & Next Steps Control Foot position tracks operator commands to within 10% under all normal load conditions Future Work Improve tracking to 5% Improve haptic feedback so that operator applies no more than 1/6 robot weight to an obstacle before detection CCEFP TB4 Brian Guerriero Conclusions & Next Steps User Interface Operator workstation in place and operational Future Work Improve AR overlays Integrate work from NCAT collaborators optimizing data feed to operator Implement sensors and tools necessary for mission success CCEFP TB4 Brian Guerriero Conclusions & Next Steps Guided-Gait Coordination Trajectory recording in place Overall algorithm ready for implementation Future Work Develop software of gait controller Develop simulation of rear four legs Integrate functions of controller CCEFP TB4 Brian Guerriero Thank You Questions? CCEFP TB4 Brian Guerriero