Robotics

1
Robotics
2
Robots

• Mobile
• Humanoid
• Legged
• Industrial

3
Sensors

• Range
• Sonar, Sick LMS, Infrared
• Inertial
• Accelerometers, Gyroscopes
• Odometry
• Shaft Encoders
• Imaging
• Camera
• GPS

4
Actuators

• Robot interaction with the world
• Motors
• Arms
• Degrees of Freedom
• Number of independent directions an actuator can
  move

5
Robot Localization

• Landmark-based Use environment to localize
• Metric-based Use robot motion to localize
• Better to use both, typically probabilistic
• Transition (Motion) model
• Sensor model

6
Robot Localization

• Kalman Filters
• Represent robot position as Guassian
• After moving, update robot position using motion
  model
• Update robot position using sensor model based on
  observation
• Disadvantage Unimodal representation

Robot Position
Landmark
7
Robot Localization

• POMDPs
• Represent world as POMDP
• Maintain robot position as belief state
• Update belief state after movement based on
  motion model
• Obtain observation
• Update belief state based on sensor model
• Advantages
• Multimodal representation
• Can be used for planning
• Disadvantages
• Requires discretization of environment
• Computationally expensive

8
Robot Localization

• Monte Carlo Localization
• Uses Particle Filters
• Keep random set of possible positions
• Update each sample based on motion and
  observation
• Select weighted sub-sample from new set
• http//www.cs.washington.edu/ai/Mobile_Robotics/mc
  l/
• Advantages
• Can handle continuous spaces
• Not computationally expensive

9
Mapping

• Evidence Grids Weighted map representation
• Updated based on sensor models
• SLAM Simultaneous localization and mapping
• Robot needs to build map and localize at same
  time
• FastSLAM
• Uses same mechanism as MCL.
• Each sample maintains map.
• http//www.informatik.uni-freiburg.de/haehnel/res
  earch/scan-matching-fastslam

10
Obstacle Avoidance

• Schema-based navigation
• Robot attracted to goal
• Robot repelled from obstacles
• Weighted sum up vectors gives resulting vector
• Generalizable to other behaviors
• GoToBall, Wander, FollowHallway, EnterDoor
• Extremely fast to calculate
• Problems with local minima

11
Architectures

• Robots contain several components which need to
  fit together
• Handle high level and low level control
• Sensing, acting, mapping, localizing, planning,
• Need to be able to react quickly

12
Reactive Systems

• Tight coupling between sensing and action
• No world state
• Sometimes represented as FSA

13
Three Layer Architecture

• Hybrid architecture
• Combines high level deliberation with low level
  control
• Reactive Layer low level controls of robot (ex
  schemas)
• Executive Layer accepts actions from
  deliberative layer and sequences actions for the
  reactive layer
• Deliberative Layer handles complex tasks such
  as planning

14
More Robots