1 of 22

ParaDocs

Team E PR 11 - Liwei

2 of 22

Progress

1

3 of 22

Our Goals for PR 11 include

  • Motion compensation integrated
  • Demonstration

4 of 22

Assumptions

  • A point of no return: after the drill is on
    • At this stage, can only stop the drilling
  • No obstacles

5 of 22

Manipulation Settings

  • Global Planner
    • OMPL Constraint Planning -> Pilz Industrial Motion Planner
    • Planning time: 4 sec -> 0.003 sec
  • Local Planner
    • FSM with execution looping
    • Compensate for small motion

6 of 22

Tracking Performance

7 of 22

Tracking <-> Manipulation

8 of 22

Additional Progress

  • Perception Pipeline Updated
    • Smarter global-registration estimate
    • Main functions optimized for speed up
    • Registration+Tracking Integrated via State Machine
  • Calibration
    • Recalibrated the intrinsics of the camera, corrected the depth estimation.
    • Re-tried extrinsic calibration with ChArUco board
    • Majority of calibration error is due to orientation difference
      • We will retry with an april tag board

9 of 22

Additional Progress: Perception

Global Estimate (Before ICP)

After 1 trial of ICP

10 of 22

Additional Progress: Perception

Before:

  • Unprojection + Preprocessing : 200ms
  • ICP : 400ms - 1.2s depending on RANSAC trials

​

​

After:

  • Unprojection + Preprocessing : 2ms
  • ICP : Single Iteration (15ms)*

*When system is not under stress

11 of 22

Demo

2

12 of 22

Backup video

13 of 22

Challenges

3

14 of 22

Challenges

  • Majority of the error left is with the calibration. Sensor noise and limitations have a role to play with how close we can get.
  • How to reduce overall load on the system.
  • Improving perception performance without affecting accuracy
  • Convergence issue due to tracking delay

15 of 22

Risk Management

4

16 of 22

Mitigated Risks

​

​

​

​

​

Risk <-> Mitigation

  • End-to-End Accuracy not spec <-> Tested
  • OMPL Constraint Planning failure <-> Switched to Pilz

17 of 22

Active Risks

​

​

​

​

​

Risk <-> Plan

  • Demonstration not clear enough <-> scope out multiple levels of demonstration

18 of 22

Future Work

5

19 of 22

Future Work

  • Incorporate drill to test end-to-end accuracy
  • Make the motion plan less jerky
  • Perception performance optimizations (potentially trying out some lighter-weight models)
  • Ironing out edge-cases

20 of 22

Thank You and Break a Leg!

21 of 22

22 of 22

Please keep this slide for attribution

CREDITS: This presentation template was created by Slidesgo, and includes icons by Flaticon and infographics & images by Freepik