Lecture

• multi-purpose
• autonomous

• exteroception – information from the outside world
• interoception – information from within the body of a robot
• proprioception – information from within related to the movement of the body parts

• LED is not an actuator :(
• effector … more complex for manipulation or mobility

• how to solve it
  • use a map (purposes of a map: localization, computing the route to a given goal)

• feasible plan = takes the capabilities of the robot into account
• motion strategies can have different forms – it can be a path, a set of procedural instructions
• for a walking humanoid robot, the plan would be quite complex

• motion planning – global map, strategic level
• obstacle avoidance – local map, tactical level

• physical space where the robot lives, usually $\mathbb R^2$ or $\mathbb R^3$
• obstacles, free space
• types of models
  • continuous metric model
    • polygons – their vertices
    • memory complexity proportional to the number of obstacles
  • discrete metric model
    • pixels: free, fully occupied, partially occupied
    • memory complexity proportional to the size of workspace and the resolution
  • topological model
    • nodes and edges
    • we don't care about the geometry – we only capture places and ways to move from one to another
    • the robot needs to be able to localize itself (what is the current “place”) and to get from one place to another according to the edges in the graph
  • hybrid
• selection criteria: sensors available, precision required, complexity (memory requirements)

• free flying robot (rigid object) = piano
• stationary obstacles (fixed rigid objects)
• the focus is on the geometry, path planning

we need to know (model) the future motions of the obstacles in order to plan the path

• two classes of kinematic constraints
  • holonomic – restrict the set of possible position (like obstacles)
    • example: limitations of the robotic arm due to its joints etc.
    • to account for the constraints, we only add some virtual obstacles
  • nonholonomic – restrict the set of possible differential motions
    • example: a wheel rolling without slipping on the plane (you should move in a direction perpendicular to the axis of the wheel)

• sources: sensors, actuators
• model error, action error

• main issues
  • safety – we don't want the robot to collide with people
  • acceptability (politeness) – the robot should not interrupt people engaging in a conversation
• attention-based HRM
  • we don't want to distract people in the museum

• mobile robot
  • reactive strategy
  • graph search, A*
  • reinforcement learning
  • robot position … coordinates
• arm robot
  • pose … vector of joint angles