DRC-HUBO: Technical Review

Jungho Lee Ph.D Rainbow Robotics, CEO

contents

1. Robot Platform: HUBO 2. Real-time OS & Framework 3. Control Strategy 4. DRC Finals

1. Robot Platform HUBO

1.0 Rainbow

Best Engineering &Technology university in Korea

World: 17th(QS), 26th(THE)

Best Robotics Research Center in Korea

1.1 Hardware Overview

History of humanoid robot, HUBO

1.1 Hardware Overview

DRC-HUBO

• Height : 155cm• Weight : 60kg

(2013)

DRC-HUBO+

• Height : 175cm• Weight : 80kg

(2015)

IMU sensor

LIDAR

Computer

Battery

Foot

Hand

P.A.C. sys.

Wheel

1.1 Hardware Overview

1.2 Light and Rigid Design

• Exoskeletal structure • Avoid cantilever • No external cables • Modular design

- Torso -

- Leg -

- Arm -

1.2 Light and Rigid Design

• No external cables - There is no external cables by using

hallow shaft

- Protect cables from malfunction and external impact

• Modular design - Facilitate assembly and repair process

1.3 Effective Heat Dissipation System

• Specially designed cooling fins with fans

- Knee joint and Hip pitch joint need much heat dissipation

- Specially designed fins absorb heat from motors and motor control boards

• Heat dissipation by using contact with frame

- Heat dissipation from motors and motor control boards to aluminum body frame

1.4 Transformable Humanoid

1.5 Robust Motor Driver

2ch and 1ch motor controllers

 

1.6 Smart Power Management

Super Capacitor

LCD Monitor

Main Controller

Li-Ion Battery 48V / 11.4 Ah

1.7 Reliable Internal Communication

PC

CAN (2ch)

isolator isolator isolator isolator

FT sensorJoint Motor Controller

Joint Motor Controller

Can High

Can Low

CAN (2ch)

(USB Connection)

Right Leg Left Leg Right Arm Left Arm

1.8 Reliable Vision/LIDAR System

PL

PM

PC

HUBO head, rotating vision sensor system HUBO head calibration

Due to rotating vision sensor system, we can obtain full 3D point cloud of target area and control laser sparsity using motor sweeping speed

2. Real-time OS & Framework

PODO-RT

2.1 How to move robots?

PODO Framework?1. “PODO” is named from Korean word “포도”, grape in English. 2. We call each process in PODO as “AL”(알), grape berry in English. 3. Many programs(processes) for controlling robots are attached to shared memory.

2.2 PODO Framework

Module 1 Library

Module 2 Library

Module n Library

Dependent Structure Multi-agent system

2.2 PODO Framework

Module 1 Process

Module 2 Process

Module n Process

Independent Structure Multi-agent system

PODO

2.2 PODO Framework

2.3 Real-time OS• “A system is said to be real-time if the correctness of a computation depends not only on the logical correctness but also on the time at which the results are produced [1].”

[1] Shin, Kang G., and Parameswaran Ramanathan. "Real-time computing: A new discipline of computer science and engineering." Proceedings of the IEEE 82.1 (1994): 6-24.

DataValidity

1

Time

deadline

Time

Data Validity

1

deadline

System Unstability

Time

deadline system failure

System Unstability

Time

deadline degrade system quality

Hard real-time

Soft real-time

missing a deadline is a total system failure.

the usefulness of a result degrades after its deadline, thereby degrading the system's quality of service.

2.3 Real-time OS

2.3 Real-time OSl Firmware

• 시스템이 간단함• Hard real-time • 실시간 연산속도에 제한 받음• 기능이 제한적임 (비 OS) • UI가 제한적임

Hard Real-time

Robot System

Firmware based Embedded System

l GPOS(Soft RTOS)

• GPOS의 기능을 활용할 수 있음• PC선택에 비 제한적임• Soft real-time 혹은 hard real-time이지만 선택적 명령 수행으로 제한됨

• 실시간 연산속도에 제한 받음

Hard Real-time

Robot System

General Purpose OS(Robot framework)

Firmware based Embedded System

Non/Soft Real-time

Communication

Hard Real-time

Robot System

General Purpose OS(Robot framework)

Communication

Hard RTOS

• Hard real-time • GPOS의 기능을 활용할 수 있음• 복잡한 연산도 가능• 비싼 가격• 시스템의 구성이 어려움• RTOS에 따라 PC가 제한적임• RTOS에 상응하는 GPOS를 쓸 수밖에 없음• Real-time 통신 모듈을 직접 구현해야 함

l Hard RTOS

2.4 PODO-RTAll the actions must start and end within one cycle of control period.

The updating time of sensor and the sending time of reference should be regular and periodic.

Time Offset(read sensor)

Time Offset(send reference)

CalculateReference

(with sensor)

Sensor #1

Sensor #N

Joint #1

Joint #2

Joint #N

Robot Hardware

Control Period(n+1)

(n)

(n-1)

(n)

(n+1)

(n-1)

(n)

(n+1)

(n-1)

(n+1)

(n-1)

(n)

(n-1)

(n)

(n+1)

Send Reference to Robot

Pass Reference to Daemon

Request ReferenceControlPeriod(5ms)

ALDaemon

WorkingTime

Suspend Time

Robot

Request Sensor Data

Generate Next Reference(Use Sensor Data)

Synchronize Reference & Sensor Data

2.4 PODO-RT

PODO ALs

Shared Memory

PODO-RT

Communication (EtherCAT, CAN, RS485, etc..)

Robot System (Controllers, sensors, etc..)

General Purpose OS (OSX, Linux, Window, etc..)

Robot Framework

PODO DaemonReal-time Kernel

3. Control Strategy

3.1 Supervisory and Autonomy

> Supervisory : Where to go and direction

Case - Movement

3.1 Supervisory and Autonomy

> Supervisory : Set Valve ROI range

Case - Task

3.1 Supervisory and Autonomy

Drill recognition Valve recognition Terrain recognition

Vision recognition result

3.1 Supervisory and Autonomy

Rotate drill to grab in correct orientation Try Several different Position and orientation to turn on the Drill

Use Mic to Detect Drill status

<Autonomy in motion : Drill task>

3.1 Supervisory and Autonomy <Autonomy in motion : Manual operation>

Auto redundancy adjust in manual control

3.1 Supervisory and Autonomy

3.2 Whole System Configuration

Robot-Motion Ubuntu 12.04 + Xenomai

i5-4250U 1.30GHz x 4

Vision-Grabbing Windows 8.1

i5-4250U 1.30GHz x 4

Vision-Field Windows 8.1

Xeon E5-1620 3.70GHz x 8

Motion-Field Ubuntu 14.04

i7-4790K 4.00GHz x 8

OCS-Main Ubuntu 14.04

i7-4790 3.60GHz x 8

OCS-Virtual Ubuntu 14.04

i7-4790 3.60GHz x 8

OCS-Monitoring Ubuntu 14.04

I7-4700MQ 2.40GHz x 8

TCP

TCP Server

UDP

CAN Bus

Robot Field

OCS

Motor Controller #1

Motor Controller #2

Motor Controller #N

Sensor #1

Sensor #2

Sensor #N

LIDAR

Camera #1

Camera #2

DRC-HUBO+

3.3 Degraded Comm. Handling

3.4 Intuitive User Interface

Monitor#1

Monitor#2

Joint Status

Program Status

Image view

Sensor info.

3D view

User button

Error signalZ-map

3.5 Compliance Control

Difficulties of force control - System is originally highly geared actuator - Harmonic drive has less back-drivability - When motor drivers on(FET ON), motor experience braking effect

- Non complementary switching mode -> Cancel braking effect - Friction compensation

-> Make back-drivable

3.5 Compliance Control

3.6 Mobility

Robot

Force, Moment, ZMP, Angle, Velocity, Vision, Etc.

Walking Motion Planner

Foot Position Foot Pose Pelvis Height Pelvis Pose

Walking Pattern Generator

CoM Whole Body Inverse Kinematics

Walking Pattern Controller

Predictive Motion Controller

Real-Time Balance Controller

Joint Angle

l Walking Framework

3.6 Mobilityl Balance Control

High precision rate gyro • Fiber Optics Gyro(FOG) • Superior bias instability (≤0.1°/hr, 1σ)

3.6 Mobility

StableC

Coliision freeC

allowable joint rangeC

l Walking Motion Planner

Valid Plane Extraction

CollisionJoint Limit

Generate Candidate Configuration

LIDAR DATA

Motion Checker

Priority Based Parameter Modification

1. Hip Height 2. Hip Rotation 3. Toe-off Motion 4. Foot Orientation 5. Foot Position

Optimized Walking Configuration

Yes

No

Search the space of stable configuration

3.6 Mobility

• Hip Height Modification Motion

• Toe Off and Pelvis Rotation Motion

l Extending walking stride

3.6 Mobility

3.6 Mobility

4. DRC Finals

4.1 Tasks

Driving

Egress

Door

Valve

Drill

Surprise

Stair

Debris

Terrain

4.2 Driving Task

4.3 Egress Task

4.4 Door Task

4.5 Valve Task

4.6 Drill Task

4.7 Unknown Task

4.8 Terrain Task

4.9 Debris Task

4.10 Stair Task

4.11 DRC Finals: TeamKAIST

4.11 DRC Finals: TeamKAIST

Prof. Junho OhDr. Jungho LeeDr. Inhyeok Kim

Dr. Jungwoo Heo

4.12 We have to do more and more…

Q&A

Thank You