5. Modeling of large state spaces
• Iden$fy key sources of uncertainty in a system
– Actua$on
– Sensing
– Communica$on
• Measure/approximate probability density
func$on
• Sample from these distribu$ons when
simula$ng the algorithm
S. Ru$shauser, N. Correll, and A. Mar$noli. Collabora$ve Coverage using a
Swarm of Networked Miniature Robots. Robo$cs & Autonomous Systems,
57(5):517‐525, 2009.
6. Example: coverage
• Algorithm
– Build a minimal
spanning‐tree on‐line
– Move from blade to
blade reac$vely
– Localiza$on by coun$ng
blades
– Start‐over when lost
• Uncertainty
– Naviga$on
8. Quan$fying Sensor & Actuator
Noise
6000 experiments in Webots, 10% wheel‐slip
Time for covering one blade Probability of no naviga$on error
(geometric distribu$on)
9. Discrete Event System Simula$on
Webots‐Generated
Event Time Data
Choose robot (closest
next event $me), add
event $me for robot
Determine next node n to visit Algorithm
Naviga$on Failure
Success? probabilites
Yes No
Move Robot Move Robot No
to n somewhere else
All Blades
inspected? Yes
13. Example: Distributed Robot Garden
• Mo$va$on: Precision
Agriculture
• Robots water and forage
tomato plants
• Pots monitor humidity
level and coordinate
robo$c system
• Robots inventory each
plant and store it into its
pot’s database
18. Filter‐based object recogni$on
• Filter image
– Sobel
– Hough transform
– Color
– Spectral
highlights
– Size and shape
Sobel Hough Color Spectral
• Weighted sum of Highlights
filters highlights
object loca$on
19. Inventory
• Challenges
– Percep$on 1 6
– Not possible from single perspec$ve
• Algorithm 2 5
- Fetch fruit inventory from pot (JSON)
- Object recogni$on from 6 non‐ 3 4
overlapping perspec$ves
- Merge observa$on with inventory
• Confidence grows with every
measurement
• Inventory dura$on: 45s
20. Visual Servoing/Grasping
• Challenges
– Percep$on (fruits + stem) 2
– Limited DOF / workspace
• Algorithm
- Select fruit with the
strongest confidence
- Servo to ini$al posi$on
- Servo to fruit using image
Jacobian
- Rely on radius es$mate for
depth F. Chaumele and S. Hutchinson, “Visual servo control
part i: Basic approaches,” Robo$cs & Automa$on
- Close gripper / retract arm Magazine, vol. 13, no. 4, pp. 82–90
when arrived
25. Possible model
T1: Harvest T2: Robot x * T3: Robot
request receives task 73s +/‐ 15s reaches plant
(p|Naviga$on failure)x
28.3s+/‐10s
25%
Assump$ons
‐ No task alloca$on (single robot) T4: Robot
‐ Infinite number of grasping trial grasps
Next step
‐simulate task alloca$on based on
communica$on model
‐finite number of fruits per plant