lri dbs

Biomimetic Sensing for Robotic Manipulation: 

Biomimetic Sensing for Robotic Manipulation Neil Petroff, Ph. D. Candidate University of Notre Dame Lerner Research Institute Cleveland, OH December 8, 2005

Outline: 

Outline Me on Me Grasping biology as motivation for current work Robotic Manipulation Nonholonomic motion planning Motion planning for stratified systems Open-Chain Manipulators Forward kinematics Inverse kinematics Biomimetic Robot Sensors Vision, touch Control Perspective on Deep Brain Stimulation The Rest of the Story

Hand Orthosis: 

Hand Orthosis Target Group: C5 - C7 SCI 3 Grasps Fingertip, key, cylindrical Increase Autonomy Mercury Orthotics Rehabilitation technology therapeutic quality of life

Grasping: 

Grasping Interaction Creation Task Execution Grasping Hand Orthosis Robotic Manipulation Fuzzy Logic Open-Chain Manipulators Biomimetic Robot Sensors Work to Date

Grasping: 

Grasping Can we improve robotic manipulation by imbuing robots with useful human characteristics? Robots Humans Poor at fine motion good at fine motion No feedback vision, proprioception structured adaptive precise robust rapid slow strong variable stamina need to rest Grasping Hand Orthosis Robotic Manipulation Fuzzy Logic Open-Chain Manipulators Biomimetic Robot Sensors Work to Date

Biological Motivation: 

Biological Motivation Haptic Recognition Force feedback Compliance is Useful for Manipulation Brain Model Fuzzy logic Hierarchical Control Grasping Hand Orthosis Robotic Manipulation Fuzzy Logic Open-Chain Manipulators Biomimetic Robot Sensors Work to Date

Biological Control Loop: 

Biological Control Loop desired task motion planning algorithm inverse kinematics encoder counts PID Robot current configuration encoder counts sensor readings trajectory adjustment fuzzy supervisor

Testbed: 

Testbed

Robotic Motion Planning: 

Robotic Motion Planning Steering Using Piecewise Constant Inputs This is a geometric analysis Provides a systematic approach for establishing controllability Applicable to underactuated systems with nonholonomic constraints Exact for nilpotent systems of the form Driftless Not all gi’s may exist a system is nilpotent if all Lie brackets greater than a certain order are zero Lie bracket motions allows the system to move in a new direction

Lie Bracket Motions: 

Lie Bracket Motions Flow along g3 can be approximated by flowing along g1 and g2 Higher order brackets can be generated, e.g.

Example: 

Example Parallel parking a car

Example: 

Example Car equations g1 g2 Extended System l

Car Simulation: 

Car Simulation

Why Didn’t it Work?: 

Why Didn’t it Work? The Car Model is not Nilpotent g5 points in the same direction as g3 Motion along lower order brackets induces motion along higher order brackets Solution Iterate Feedback nilpotentization Other Drawbacks Small Time or Small Inputs obstacle avoidance Open Loop highly susceptible to modeling errors no error correction

Stratified Systems: 

Stratified Systems Extends motion planning algorithm to systems with discontinuities Intermittent contact locomotion manipulation

Control Architecture: 

Control Architecture Desired task motion planning algorithm

Open-Chain Manipulators: 

Open-Chain Manipulators Forward kinematics s P T

Inverse Kinematics: 

Inverse Kinematics The inverse kinematics solution is not unique 1 1 1 1

Inverse Kinematics: 

Inverse Kinematics PUMA geometry makes an analytical solution tractable

Inverse Kinematics: 

Inverse Kinematics 14” diameter circle

Control Architecture: 

Control Architecture Desired task motion planning algorithm inverse kinematics encoder counts PID Robot current configuration current counts fuzzy supervisor

Biomimetic Sensing: 

Biomimetic Sensing

Force Sensors: 

Force Sensors Feedback at Finger/Object Junction Piezoelectric Used in biomedical testing Compliant Tend to drift under static load Flexiforce Sensor

Finding an Object: 

Finding an Object

Control Architecture: 

Control Architecture desired task motion planning algorithm inverse kinematics encoder counts PID Robot current configuration encoder counts sensor readings fuzzy supervisor trajectory adjustment

Summary: 

Summary So Far Built a closed loop system to perform robotic manipulation stratified motion planning inverse kinematics solution force feedback To Do Manipulation Currently working on simulation apply to robots

Control Perspective on DBS(or “What the heck am I doing here?”): 

Control Perspective on DBS (or “What the heck am I doing here?”) Underlying manipulation technique is a geometric approach to nonlinear controls Nonlinear control lies at the forefront of modern control methods One of the most intriguing aspects of nonlinearity is that of chaos Nonlinear control techniques have been used to suppress cardiac arrythmia, a chaotic process Is neuron transmission chaotic? at the heart of successful treatments using deep brain stimulation is the ability to control chaos Robust and nonlinear control techniques provide an analytical foundation on which to study such systems Soft computing techniques provide an additional approach that while not at rigorous may yield equal or better results

Open Questions on DBS: 

Open Questions on DBS By approaching DBS from a control Theory Standpoint, Can We Control with external stimulation locally? Filter the signals? Characterize which signals cause which disruptions stimulation can suppress dyskinesia tremors tend to lessen during movement Keep symptoms from returning with fatique? Muscle spasticity Completely eliminate meds?

The Rest of the Story: 

The Rest of the Story 54,000 SCI Additional 2,800 / yr at C5 – C6 level Parkinson’s affects 750,000 – 1 million people in the U.S. Other Pathologies Hemiplegic stroke Multiple sclerosis Muscular dystrophy Rehab Funding Competition for startup money Who Can Pay? Hand Mentor from KMI $3,950 Coverage from private insurance companies in only 2 states Currently no medicare coverage State of Indiana Home and Community Based Care Act Provides funding for community and home-based care 2002: 84 / 16 Medicaid savings of $1,300 per client per month Savings on the order of 3:1 when compared with institutional care

My Plea: 

My Plea As researchers, I believe we have a responsibility to pursue noble goals Obligation of the Engineer “… conscious always that my skill caries with it the obligation to serve humanity …” Hippocratic Oath “I will remember that I do not treat a fever chart, a cancerous growth, but a sick human being, whose illness may affect the person's family and economic stability. My responsibility includes these related problems, if I am to care adequately for the sick.” “will remember that I remain a member of society, with special obligations to all my fellow human beings, those sound of mind and body as well as the infirm.”

On a Lighter Note: 

On a Lighter Note

Motion planning algorithm: 

Motion planning algorithm Solve for v’s from desired trajectory Expand vector exponentials and equate coefficients Solve for h’s by equating B’s of above

3rd order bracket: 

3rd order bracket

Fictitious Input Flow: 

Fictitious Input Flow

Revolute Joint Lemmas: 

Revolute Joint Lemmas Position Preservation Distance Preservation

Stratified Motion Planning: 

Stratified Motion Planning If t4 = t6 and t1 = t3, Motion planning performed on S12 with projected vector fields

Contact Coordinates: 

Contact Coordinates Mapping from R3gR2 Shows evolution of finger on object EOMs on the sheets

Grasp Constraints: 

Grasp Constraints End effector motion is limited due to contact with object Present control system such that it is in “standard” form Relative contact velocities are control inputs Defines joint torques

Extended System: 

Extended System Motion planning for smooth systems (extended) The vis are fictitious inputs for extended system, pick trajectory, Can write any flow:

Lie Bracket: 

Lie Bracket Given two vector fields, g1 and g2, we can generate a third which points in a new direction This is an approximation by TSE Higher order brackets can be generated, e.g.

Inverse Kinematics: 

Inverse Kinematics Finding

Two twists: 

Two twists

Slide43: 

Orthosis Design and Feedback Requires joint angle feedback - difficult Change in DIP Angle of the Second Metacarpal of the Right Hand for Three Subjects During Flexion (Both Angles are Relative to the MCP)

Slide44: 

Measuring the MCP Angle

Mercury Orthotics: 

Mercury Orthotics 2004-2005 Notre Dame Business Plan Competition Semi-Finalist Mission: Provide state-of-the art rehabilitation technology for therapeutic and quality-of-life aid Problem I: Rehabilitation of Hand Injuries Current structure of therapy is restrictive facility based requires dedicated therapist Problem II: Assistive Aid for Long-Term Care Current techniques are invasive or incomplete Solution: HandStand System Provides a method to automate certain therapy functions facility or home based Automatic storage of relevant information and progress assessment therapist is freed to focus on patient care while treating more patients Provides a noninvasive method for restoring basic hand function responds to user commands

Phillip Hall Basis: 

Phillip Hall Basis Forms a basis for a Lie algebra A basis is like spice? Vector fields are elements of the algebra Some of the vector fields created by Lie bracket operations are in this basis These generate new directions in which a system can move Once we have enough to span the space, any point is reachable from any other point

General Solution Approach: 

Determine the kinematic equations of motion, velocity constraints Determine the vector fields which annihilate the constraints Directions in which the system is able to move Determine the Phillip Hall basis Eliminate additional, linearly dependent vector fields Describe the “extended” system comprised of actual inputs and fictitious inputs generated by Lie bracket motions Define a nominal steering trajectory Determine the inputs The span of the set of remaining linearly independent vector fields determines the involutive closure, The dimension of equals the dimension of the configuration space meaning the system is small time, locally controllable Since the distribution is involutive it can be integrated General Solution Approach

Configuration Space: 

Configuration Space Partition of M into submanifolds Different EOMs on each stratum Restricted to each stratum - equations are smooth Cyclic strata switches manipulation Consider the sequence

Fuzzy Logic: 

Fuzzy Logic Mamdani Inference System