Khatib: Realtime Obstacle Avoidance for Manipulators and Mobile Robots

1
Khatib Real-time Obstacle Avoidance for
Manipulators and Mobile Robots

• Jeffrey Yu

2
Plan

• Review the motion problem
• Show how previous approaches are similar
• Show how the potential field approach is
  different
• Describe details of the potential field approach
• Summarize strengths and weaknesses

3
Review Motion Planning Problem

• Move a robot from a start position to an end
  position
• Avoid obstacles along the way
• Complete the task in a reasonable amount of time
• Previous approaches roadmap, cell decomposition,
  probabilistic roadmap

4
Previous Approaches Common Characteristics

• Find a valid path to the goal
• Transform valid path into forces of motion
• Apply forces to the robot

5
Potential Field Approach

• Design a force to guide the robot towards the
  goal
• Design forces to guide the robot away from
  obstacles
• Apply the total force to the robot

6
Example

• Combination of forces move steer the robot around
  a corner

7
Goal Force Design Criteria

• Move the robot towards the goal state
• Bring the robot to rest at the goal state
• Stay within robot safety and operational limits

8
Goal Force 1

• Basic spring-like force
• Stable at the goal
• Weakness cyclic motion away from the goal

9
Goal Force 2

• Drag term brings the robot to rest
• Weakness long distances produce high
  acceleration and high-velocity motion

10
Goal Force 3 (Final)

• Based on goal force 2
• Produces straight-line, velocity-limited movement
  toward the goal
• Motion perturbed near obstacles

11
Obstacle Force Design Criteria

• Steer robots away from an obstacles surface
• Influence is negligible away from an obstacles
  surface
• Two methods analytic obstacle equations and
  shortest distance to an obstacle

12
Obstacle Forces Analytic Equations

• Function f(x) 0 describes obstacle surface
• Influence starts at f(x)x0, approaches infinity
  at f(x)0
• No distance calculation needed

13
Obstacle Forces Shortest Distance

• Produces motion away from the closest point of
  danger
• Real-time sensing possible

14
Strengths

• Calculations done in operational space avoid
  translation to and from configuration space
• Calculations done in control domain no
  conversion from intermediate path representation
  to command forces
• Less upfront effort expended planning a possibly
  obsolete path

15
Weaknesses

• Failure due possible even if a valid path exists
• Workarounds
• Random walks
• Roadmap connecting local minima
• Motion attempted even if no valid path exists
• Failure modes are observable to bystanders

16
Other Topics

• Manipulator optimizations
• End-effector decoupling parameters can be
  computed at a lower rate than the end-effector
  force vector
• Applications
• Khatib et al 1984 PUMA 560 with machine
  intelligence Corp vision module
• Khatib, Burdick, and Armstrong 1985 real-time
  end-effector motion with active force control

17
Summary Potential Field Methods

• Artificial forces guide robot motion
• Sacrifices reliability for speed