Progress Review 9

MRSLAM

  • Testing and tuning in real life
  • Initial testing with 4 robots – identified initialization issue
  • Modified Ray Casting behaviour to fill in all free areas

Navigation

  • Hardware testing for Lazy Traffic Controller
  • Tune the controller velocity calculation frequency to the execution frequency of the robots
  • Since differential drive robots don’t immediately achieve commanded velocity, use feedback velocity (calculated from change in position of the robot) for velocity obstacles calculation
  • Addition of new term in cost function while scoring velocities to reduce oscillations
    • float penalty = RVO_SAFETY_FACTOR / min_t_to_collision + dist_to_pref_vel + dist_to_cur_vel
  • Static obstacle avoidance
    • Latencies in real life / localization errors causes robots to sometimes collide with obstacles
    • Adding static obstacle avoidance increases robustnesses and prevents crashes

Task Generation

  • Frontier Exploration
    • RRT-based frontier generation
    • RRT expands from robot starting position into free space
    • When a node connects into unknown space, it is considered a frontier
    • Possible frontiers are then clustered and filtered to prevent redundancies
  • Challenges:
    • Improve filtering based on information gain
    • RRT does not update with map loop closures
    • How does this scale to larger maps?