DARPA Investments in GEO Robotics Carl Glen Henshaw, Ph.D. Signe Redfield, Ph.D. Naval Center for Space Technology U.S. Naval Research Laboratory Washington, DC 20375 May 22, 2015
Introduction Program Overview
DARPA Space Robotics Programs 2003 Present SUMO (2003 2005) FREND (2005 2010) Phoenix (2010 2014) Robotic Servicing of Geostationary Satellites (RSGS) (2014 present)
DARPA Space Robotics Since 2003 Artist s Concept Artist s Concept
NRL Test Facilities (left) Proximity Operations Testbed (POT); (right) Contact Dynamics Testbed (CDT)
Robotic Control Algorithm Stack for RSGS
Robot Control Modes for RSGS scripting the robotic arm moves through a pre-planned trajectory using only proprioceptive sensors. partial autonomy the robotic arm carries out a task using environmental sensors such as end effector cameras or the force-torque sensor. Tasks are carried out in steps, with the human operator explicitly issuing authority to proceed after each step is completed. full autonomy the robotic arm carries out a task using environmental sensors, detects completion of steps, and automatically proceeds under human supervision if the last step was successful. teleoperation motion of the robotic arm is commanded directly by a human operator using a hand controller, usually as commanded end effector motions. The robotic arm executes onboard servo control, and compliance control as needed, but all other control loops are closed via the human operator s eyes and hands.
Validation and Verification (V&V) Philosophy for RSGS Use the right tool for the right job Formal methods and theoretical analysis where appropriate Brute force analysis where necessary and possible Otherwise: Generate accurate dynamic models of low level components (servo loops, compliance, machine vision feature detectors build dedicated test fixtures as appropriate perform theoretical performance analysis where possible to understand stability margins validate analysis on test fixtures Incorporate lower fidelity versions into more encompassing models Identify edge cases via Monte Carlo simulation Verify performance on edge cases in POT/CDT with hardware in the loop testing
Servo Control Consists of joint and motor dynamics plus linear control law plus feedforward Newton Euler linearization plus friction compensation To validate: construct single actuator test rig generate accurate model of joint dynamics test at temperature test at various loadings perform linear gain/phase margin analysis perform Lyapunov analysis of tracking performance with nonlinear feedforward terms verify performance on test rig post tuning generate probabilistic friction model (needed later)
Compliance Control C m θ = C p θ C d θ + τ where θ is a modification of the nominal reference trajectory θ r (t) C p, C d, and C m are the desired stiffness, damping, and virtual inertia matrices τ is the vector of joint torques as calculated by τ = J T (θ) f V&V by: using model ID from above, perform theoretical analysis of stability margin during contact vary latency and loop rate to determine stability margin validate experimentally using single actuator test rig revise gains in light of stability margins develop simplified model of exerted forces and torques during contact
Machine vision detectors and visual servoing build dedicated hardware testbed with flight cameras and lights, solar simulator, and metrology system develop photorealistic simulation of spacecraft structure and thermal blankets under orbital lighting use both human analysis and Monte Carlo runs of photorealistic simulation to identify and test edge cases develop a statistical model of machine vision performance in a select number of cases
Inverse kinematics and collision avoidance We use Resolved Motion Rate Control (RMRC): θ = J + (θ) (ẋ d + λ (x d x)) (1) with J + ( ) calculated (as in English et al) to optimally choose a nullspace projection that minimizes a metric of our choosing, typically trading off collision avoidance and singularity avoidance. V&V of the IK algorithm also constitutes verifying that the kinematics are correct, that the environmental collision model is correct, and that the optimization metric is correct. using simplified version of servo control, compliance control, and machine vision models, develop a full-up dynamical sim of the robotics payload based on a scientific grade physics engine implement front end to physics engine identical to the servo/compliance control APIs perform Monte Carlo simulation of IK and collision avoidance; use automatic failure detection to flag edge cases
Mission sequencer/fdir Design the fault response decision logic using a formal language spec, which will rigorously guarantee that the flight logic is implemented correctly Use brute force techniques to verify the correct functionality of the streamed, command line, and sequenced internal control modes Perform Monte Carlo simulation of individual robotic operations and sequences using full up sim and built in automatic failure detection to flag edge cases
Integrated system Perform Monte Carlo sim of samples of the operational space to identify edge cases resulting from subsystem integration Perform Monte Carlo sim centered around nominal cases and edge cases identified by previous subsystem testing Perform POT and CDT based hardware testing of selected nominal and edge cases On orbit verification
On Orbit Verification Verify communications between flight computer and arm servo control electronics Release launch locks Perform scripted launch lock arm release trajectories (verifies baseline performance of the arm servo control and ensures the automatic sequence control flow is operational) Perform joint by joint scripted maneuvers through each joint s range of motion Collect joint tracking data Re verify joint dynamics model verifies command line control flow element in mission sequencer is working Perform collision free Cartesian motions Verifies functionality of RMRC Perform visual inspection of payload deck Teleoperated trajectory (verifies teleoperation control flow is working)
On Orbit Verification, cont. Calibrate end effector force/torque sensor Force/torque sensor must be re tared at to be determined intervals; bias values provide insight into whether sensor is working correctly Analyze sensor noise Re-verify force/torque sensor model Open/close tool changer Engage the camera calibration fixture Constitutes the first contact operation with the robot arm; acts as a validation of the compliance control system Also constitutes the first operation of the tool changer against a tool changer interface Take camera imagery of calibration targets Calculate camera light path model parameters Toggle camera fill lights Disengage with camera calibration fixture
On Orbit Verification, cont. Tool checkout: Engage with each tool and remove from its docking fixture Operate tool through range of motion Return tool to fixture (These actions will encompass varying levels of operator interaction, using partial autonomy and eventually automated command sequences to verify that the operator control modes and the sequence control flow element are working properly)
Conclusion V&V for RSGS is nontrivial No one tool is sufficient V&V strictly using hardware testing is infeasible due to high number of conditions Rigorous tools available for important subsystems (servo control, compliance control, IK) Make use of computing power via Monte Carlo sims to identify cases of interest Make use of ground test facilities and early flight hardware availability to validate in these cases The views, opinions, and/or findings contained in this presentation are those of the presenter(s) and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.