Impact

1
Skill Learning by Primitives-Based Demonstration
Imitation

• New Ideas
• Low cost, lightweight, wireless, real-time
  human motion collection
• Cross-kinematic analysis for learning motion
  primitives from various sources
• Creating controllers from collected sensory
  data
• Randomized dynamic roadmaps for humanoid
  reaching and for planning primitive motion
  sequencing

Motion controllers based on randomized roadmaps
and collected sensory/motion data
USC motion suit
A Cross-Kinematics Metric for Imitation Learning
Robonaut Control
Accomplishments
Impact A methodology for on-line human-robot
control Parameterized controllers allow
human-robot cooperation for a variety of tasks
Increased autonomy for humanoid robots
Teleoperation of NASA Robonaut
Integrated demo in Robosim
Sensory data analysis
USC motion suit
Year 1
Year 2
Cross-Kinematic Metric
Reaching Motion Planning
USC humanoid/Robonaut test-beds
Motion analysis from Robonaut Data
PI M. Mataric, USC