Reactive and Adaptive Control Strategies for Robot Assembly

Initial Infrastructure updates

Aditya Kamireddypalli

Sprint 1

Simulation

• Drake and MujoCo
• State Estimation loop
• Motion planning
  • Each action as a controller
  • Base Impedance Control
  • Traj Opt 

Real Robot

• Peg-insert on existing gear setup, Panda
• State Estimation Loop
• Motion Planning
  • Impedance Control set-point

Sprint 1 - retrospective

Simulation

• Drake Peg-insertion Finite State Machine
  • State Estimation loop
    • Particle Filter - doubtful implementation.
  • Motion planning
    • Set point control of Stiffness Controller
  • Symbolic State Estimation
    • Termination Predicates
    • Time based, local disk based (squared error)
• Real Robot
  • Experimental infra design
  • No real robot experiments yet

Sprint 2

Simulation

• Polish the low level State estimation loop
  • Experiment with GTSAM and other code bases
• Motion planning
  • Set point Stiffness Control
  • Traj Opt 
• Predicate engineering

Real Robot

• Peg-insert on existing gear setup, Panda
• Motion Planning
  • All sim controllers can be used with Real Robot, courtesy drake_ros

Sprint 2

Simulation

Predicate Engineering

• Simple Force based classifiers - baselines
• Composition of multiple predicates. Gaussian Mixture Models
• Action sensors
  • Model based controllers
  • Use feedback and construct local disks around EEF. Use Intersection between ideal disk and estimated disk as a sensor.
  • Propagate this as a component into the Gaussian Mixture

Stretch

• Connect up the software with real robot and demonstrate.