What did we learn from Darpa Robotics Challenge

Transcription

1 Keynote speech/icra, Hamburg, Germany, Sep What did we learn from Darpa Robotics Challenge Jun Ho Oh Professor of Mechanical Engineering Director of Humanoid Robot Research Center KAIST

2 DARPA Robotics Challenge Supervision Defense Advanced Research Project Agency DARPA Purposes To generate groundbreaking research and development so that future robotics can perform the most hazardous activities in future disaster(like Fukushima disaster) response operations, in tandem with their human counterparts, in order to reduce casualties, avoid further destruction, and save lives. 2

3 DARPA Robotics Challenge Drive a utility vehicle at the site Egress from the vehicle Travel across over the rubble or get through.

4 DARPA Robotics Challenge Open a door and enter a building. Climbing stiff stairs.

5 DARPA Robotics Challenge Use a tool to break through a wall. Locate and close a valve near a leaking. Execute surprise task. 5

6 100Mb/s 9600 b/s

7 DARPA Robotics Challenge Jaxon (Univ. of Tokyo) HRP+ (AIST) Chimp (Carnegie Mellon University) Atlas Walk-Man (IIT) (Boston Dynamics & DARPA) 7 Robosimian (NASA)

8 Stair Debris Terrain Surprise Drill Valve Driving Door Egress

9 Fairplex, Pomona, California

10 DARPA Robotics Challenge

11

12 DARPA Robotics Challenge l We have to do more and more. 12

13 What happened? Disaster response robot encountered real disaster before getting into the disaster scene by itself.

14 Cause of Failures System and hardware failures Insecure mobility(biped walking) Poor interactivity with environment Unbalanced control between autonomy and supervisory Lack of expertise, confidence and practice in operation Operator mistakes

15 l Winning Strategy Robust Robot Platform Robust mechanism No external cables Stiff mechanical structure Modular design Smart power management Stable power for PC, communication, etc. via Super-capacitor Soft start-up circuits Effective heat dissipation Custom designed finned air cooling with a fan Heat dissipation through body Reliable internal communication Isolators 4 CH-1Mbps CAN Robust motor driver 15

16 l Winning Strategy Reliable vision/lidar system No Kinect RGBD sensor No stereo vision 2D vision plus LIDAR Custom designed LIDAR scanner Balance between Supervisory and Autonomy Supervisory: Indicate Where to go in 2D vision space. Indicate Region of Interest (ROI) for the objects in 2D. Autonomy: Get objects in 3D cloud data from LIDAR Script base autonomy Manual: Augmented virtual space manipulation in 3D cloud data Visual servo 16

17 l Winning Strategy Real time OS and framework for multi-agent system BotRu Stable and reliable mobility Transformable to upright and kneeling postures Compliance control Dynamic egress - Computed torque method Door, Valve, etc. Variable low gain Realistic and comprehensive rehearsal Practice under various sun light and wind gust conditions Practice without any safety gear over rough ground OCS room 200m apart from the sight Practice under 2.4k bps bandwidth condition Leadership and team-work 17

18 DRC Hubo 18

19 2ch and 1ch motor controllers 2ch BLDC or DC motor controller with amp. Up to 48V motor voltage (200Watt for each channel) Microchip dspic (40 MIPS) 2ch encoder and hall sensor input, CAN and 2 DIO Temperature & current sensing State indicator using seven-segment Auto scaling algorithm based on main control loop time Functions for setting parameters (gain, vel. and acc. limit, home position, limit position and so on ) Various error control algorithm (big error, encoder failure, over current and so on) 90mm 80mm 20mm 19

20 LCD Monitor Super Capacitor Li-Ion Battery 48V / 11.4 Ah Main Controller 20

21 DARPA Robotics Challenge Driving Utility Vehicle 21

22 DARPA Robotics Challenge l Egress Task 22

23 DARPA Robotics Challenge Stabilization

24 24

25 Case - Movement > Supervisory : Where to go and direction 25

26 Vision recognition result Drill recognition Valve recognition Terrain recognition 26

27 Operating System - Hyper number of multi-agent - Composite movement - Strict synchronization Plus - Multi programmers - Easy programming w/o pre-knowledge - Light

28 l Real-time OS: BotRux - BotRux is named from Latin Botrus means grape. - We call each process in BotRux as AL ( 알 ), grape berry in English. - Many programs for controlling robots are attached to shared memory. 28

29 Very Realistic Practice and Rehearsal Practice Area OCS Room

30 DARPA Robotics Challenge Thank you