Collaborative Mobile Robots for HighRisk Urban Missions Report on Timeline, Activities, and Mileston

1
Collaborative Mobile Robots forHigh-Risk Urban
MissionsReport on Timeline, Activities, and
Milestones

• P. I.s Leonidas J. Guibas and Jean-Claude
  Latombe
• Computer Science Department
• Stanford University
• http//underdog.stanford.edu/tmr
• December 15, 1998
• DARPA TMR Program

2
Timeline for Model Building
3
Timeline for Target Finding
4
Timeline for Target Tracking
5
Q2 - Map building Next-Best View Techniques

• Explanation Next-best view technique for
  cooperative map-building with multiple robots
  (simulation).
• Benefits
• 2D map is used for deciding robot positions for
  complex sensing operations (Q4).
• 2D map can be used as workspace for target
  finding and tracking (Q3-Q7).
• Cost 64,225.

6
Q3 - Map building localization uncertainty

• Explanation Next-best view technique for map
  building of interior terrain with real robot
  under large localization uncertainty.
• Benefit
• Makes next-best view algorithm more robust for
  practical environments.
• Generates more precise 2D maps for other tasks.
• Cost 66,500

7
Q3 - Target finding extended sensory models

• Explanation Simulation of target finding planner
  with new sensory model.
• Benefit
• Generates reliable motion plan that guarantees
  target cannot escape into already swept region.
• Works even under restricted sensing conditions
  (e.g. cone of vision).
• Used in Q5 for target finding techniques when
  there are not enough robots.
• Cost 64,105.

8
Q4 - Map building art gallery techniques

• Explanation Randomized art-gallery technique to
  find a minimized set of locations allowing a
  camera-equipped robot to take pictures covering
  an entire environment.
• Benefit
• Exploits 2D map constructed in Q2 and Q3 to
  compute robot positions.
• Minimizes number of necessary complex sensing
  operations.
• Cost 61,950.

9
Q4 - Target Tracking One Robot, One Target

• Explanation Target-tracking planner for one
  robot and one target, using a joystick robot as
  target.
• Benefit
• Important step towards general goal of multiple
  robots tracking multiple targets (Q5-Q7).
• Demonstrates key ideas used in target tracking.
• Cost 47,100.

10
Q5 - Target finding non-guaranteed strategies

• Explanation Target finding strategy that
  minimizes the amount of space in which the
  targets may still hide, when there are not enough
  robots to reliably find all targets
  (simulation).
• Benefit
• Guarantees targets are hiding in as small an area
  as possible.
• Extends target finding technique when there are
  enough robots (developed in Q3).
• Cost 64,950

11
Q5 - Target finding with communication maintenance

• Explanation Target-finding technique where the
  robots maintain a communication network.
• Benefit
• Robots protect each other.
• Robots can share information through
  communication channels.
• Cost The cost of this milestone is integrated
  into the costs of the other target-finding
  milestones.
• Note This is an additional accomplishment that
  is not in the contract.

12
Q5 - Target Tracking Several Robots, Several
Targets

• Explanation Target-tracking planner allowing N
  robots to track Q robots, with N and Q roughly
  the same (simulation only).
• Benefit
• Allows multiple robots to track multiple targets.
  
• Permits robots to exchange tracked targets.
• Cost 46,500.

13
Q6 - Target Finding One Robot

• Explanation Demonstration of target-finding
  capability with one robot in a real indoor
  environment.
• Benefit
• Demonstrates effectiveness of target-finding
  technique developed in Q3.
• Essential step towards demonstration of target
  finding with two robots in Q7.
• Cost 63,800.

14
Q6 - Target Tracking Two Robots, One Target

• Explanation Target-tracking planner for two
  robots and one human target. If required,
  perception will be simplified by having the
  target carry an easily distinguishable pattern.
• Benefit
• Demonstrates how multiple robots can coordinate
  to track one target.
• Uses target tracking algorithm for several robots
  and several targets (Q5).
• Cost 46,700.

15
Q6 - Real-time path planning

• Explanation Planner that computes path in real
  time even in the presence of moving obstacles.
• Benefit
• Can be incorporated in all planning techniques
  used for map building, target finding, and target
  tracking.
• Real-time property does not affect speed of these
  techniques.
• Cost We hope to be able to support this
  milestone with separate funding (NSF fellowship).
• Note This is an additional accomplishment that
  is not in the contract.

16
Q7 - Target Finding Two or More Robots

• Explanation Target-finding capability with two
  or more robots in a real indoor environment.
• Benefit
• Demonstrates applicability of target-finding
  technique developed in Q3.
• Is used to find targets in more complex buildings
  than earlier target finder (demonstrated in Q6).
• Cost 63,585.

17
Q7 - Target Tracking Two or More Robots and
Targets

• Explanation Target-tracking software for two (or
  more) robots and two (or more) human targets.
• Benefit
• Demonstrates technique for multiple robots
  tracking multiple targets.
• Allows robots to exchange targets being tracked.
• Cost 48,100.
• Accomplishments leveraged Robot controller that
  adjusts for delay (over network).

18
Q8/9 - Integrated Demonstration

• Explanation Demonstration of an integrated
  approach to implementing map building, target
  finding, and target-tracking capabilities in a
  government-approved scenario.
• Benefit
• Final integration of techniques developed in the
  contract.
• Final documentation and reporting of
  accomplishments during the TMR program.
• Cost 112,485.

19
Financial Summary

• Changes to original contract none.
• Total funded 188,000.
• Amount spent (as of 11/30/98) 111,716.
• Amount available (as of 12/1/98) 76,284.
• We expect the current funding to enable us to
  continue the research through 1/31/99.

20
Interaction with other TMR Contractors

• Kurt Konolige (SRI) Discussions to enable our
  robots to have omnidirectional vision by
  equipping them with SRIs spherical mirror.
• Greg Hager (Yale) Discussions to use Yales
  visual tracking algorithms in our motion-planning
  algorithms.