||Péter Fankhauser 27.02.2017

Locomotion Concepts

Spring 2017

Péter Fankhauser, Marco HutterRoland Siegwart, Margarita Chli, Martin Rufli

1Autonomous Systems Lab

Locomotion Concepts Principles Found in Nature

§ Nature came up with a multitude of locomotion concepts

§ Concepts found in nature§ Difficult to imitate technically§ Do not employ wheels§ Sometimes imitate wheels (bipedal

walking)

2||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Locomotion Concepts Principles Found in Nature

§ Nature came up with a multitude of locomotion concepts

§ Concepts found in nature§ Difficult to imitate technically§ Do not employ wheels§ Sometimes imitate wheels (bipedal

walking)

§ Most technical systems today use wheels or caterpillars

2||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Locomotion Concepts Walking vs. Rolling

3||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Legged systems can overcome many obstacles

Locomotion Concepts Walking vs. Rolling

3||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Legged systems can overcome many obstacles

§ But it is hard to achieve this:§ Floating base (vs. fixed-base)§ Many DOFs must be controlled in a

coordinated way§ The robot must interact with (uncertain)

terrain

History of Legged RoboticsWalking Mechanism – First Patents

4||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Davis, 1878Rygg, 1893

History of Legged RoboticsWalking Mechanisms

5||Péter Fankhauser 27.02.2017Autonomous Systems Lab

General Electric’s WalkingTruck, ca. 1965

Shigley's WalkingMachines, 1957–60

History of Legged RoboticsWalking Mechanisms

6||Péter Fankhauser 27.02.2017Autonomous Systems Lab

OSU ASV (Adaptive Suspension Vehicle), 1984

History of Legged RoboticsLarge Scale Legged Locomotion and Manipulation

7||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Menzi Muck

History of Legged Robotics Phony Pony, GE Hardiman and many more…

8||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ More on http://cyberneticzoo.com § steam-actuated humans § mechanical elephants § …

History of Legged Robotics Humanoid Robots After 2000

9||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Honda Asimo Toyota Humanoid

History of Legged Robotics Humanoid Robots After 2000

10||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Fukushima, 2011 DARPA Robotics Challenge, 2012

DARPA Robotics Challenge … and a thing we learned after the DRC Finals

12||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Walking is still difficult

Modeling Legged Locomotion Inverted Pendulum

13||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Walking can be represented by an inverted pendulum

Modeling Legged Locomotion Passive Dynamic Walker

14||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Exploit the inverted pendulum character in passive dynamic walkers

§ Forward falling combined with passive leg swing

§ Energetically very efficient

Modeling Legged Locomotion Passive Dynamic Walker

15||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Add little actuation to walk on flat ground

Cornell Ranger, 2011 Total distance 65.24 km Total time 30:49:02 Power 16.0 WCOT 0.28

Modeling Legged Locomotion Cost of Transport (COT)

16||Péter Fankhauser 27.02.2017Autonomous Systems Lab

usedE m g h hCOTm g d m g d d

× ×= = =

× × × ×

Modeling Legged Locomotion Energy Expenditure for Different Gaits

17||Péter Fankhauser 27.02.2017Autonomous Systems Lab

2331Mechanics and energetics of horse locomotion

analyse the three-dimensional data and to calculate thebiomechanical variables for 209 strides.

Metabolic measurementsStandard open-circuit respirometry was used to determine

the metabolic power requirements. The expired gas, flowingthrough a mask firmly positioned around the nostrils (BRDL,Birmingham, UK), was collected in Douglas bags, andventilation rate was assessed by using a chronometer and a drygas meter (Collins, USA). O2 and CO2 partial pressures weremeasured using gas analysers (Binos 1, Leybold-Heraeus andOxynos 100, Rosemount, Germany). After converting valuesto STPD (laboratory air temperature was always within the range14–16 °C), the rate of O2 uptake was calculated. Values for thestanding horse were subtracted from the steady-state values,reached after approximately 4 min of exercise, to obtain the netmetabolic work rate during each trial. To express the metabolicwork (C) in the same units as mechanical work (J kg−1 m−1),we divided the net metabolic work rate by the speed (m s−1)and assumed that 1 ml of O2 is equivalent to 20.1 J (for a meanrespiratory quotient of approximately 0.95) to convertmetabolic into mechanical units.

The efficiency of locomotion, the ratio of mechanical tometabolic work, was calculated by dividing WTOT by C.

ResultsThe rate of energy consumption of standing horses was

1.94±0.48 ml O2 kg−1 min−1 (mean ± S.D., N=4). Fig. 2 shows,for each gait, the metabolic cost at different speeds. Becauseof technical problems related to air leakage through the mask,only a subset of all the measurements was considered. Themain criteria for rejecting some measurements were the lackof an increase in ventilation with increasing speed (within eachgait) and a respiratory quotient exceeding 1.0.

Typical changes in mechanical energy for walking, trottingand galloping are shown in Fig. 3. By summing all the energyincreases in ET over the stride period and expressing the resultper unit mass and distance, we obtained WEXT. Fig. 4 showsthese values (upper curve) and WINT (lower curve).

Table 1 reports the regression coefficients for parabolicregressions for the main variables discussed in the paper.Second-order regression was used for descriptive purposes,since it can fit most linear and non-linear data without furtherincreasing the complexity of the analysis.

DiscussionThe energetics of horses locomotion shows some

optimisation phenomena (Alexander, 1989). At each gait, aspeed exists at which C is minimal, as originally discovered ina study on small horses (Hoyt and Taylor, 1981). In addition,minima in C are comparable at the different gaits, showing anoverall constancy of the metabolic energy spent per unitdistance travelled, despite the difference in speed, although thevalue of C during walking is lower than that reported by Hoyt

and Taylor (1981). This is because the subtraction of basal(standing) metabolic rate from the metabolic measurements, asapplied here, has a greater effect at low speeds than at higherspeeds (for a review of different techniques for calculating C,see Full, 1991). While exceeding (by approximately 30 %) theallometric equation obtained for mammalian and bird specieswith masses ranging from 0.01 to 254 kg (Taylor et al., 1982),the present values of C are comparable with previousmeasurements on horses (Farley and Taylor, 1991; Eaton et al.,1995; Potard et al., 1998; Langsetmo et al., 1997; Butler et al.,1993; Wagner et al., 1989).

Mechanical workThe mechanical external work (Fig. 4, upper curves) during

walking and trotting resemble the values for humans atcorresponding gaits (Cavagna et al., 1976). Fig. 5 shows thethree-dimensional range of movement of the centre of mass ofthe horse during single strides at the different gaits and speeds.It is surprising how the lateral displacement (dz) for a 500 kghorse drops from 50 mm (slow walking) to only a fewmillimetres during fast galloping and how the verticaldisplacement (dy) more than doubles when passing fromtrotting to galloping.

As it is calculated (Cavagna et al., 1976), WEXT representsa good approximation of the mechanical work done by themuscles only for non-bouncing gaits such as walking. During

Fig. 2. Relationship between the metabolic energy cost (per unitmass and distance) and speed during walking (W), running (R) andgalloping (G). The thick curves refer to parabolic fits of theexperimental data (the regression coefficients are given in Table 1),while the thin curves represent the parabolic fits of the data of Hoytand Taylor (1981) for a 140 kg horse. Filled circles shows meangalloping cost and bars refer to standard deviations (S.D.) of datataken from the literature (Eaton et al., 1995; Potard et al., 1998;Langsetmo et al., 1997; Butler et al., 1993; Wagner et al., 1989).Open circles represent data for walking, filled squares for trottingand open diamonds for galloping.

0

0.05

0.10

0.15

0.20

Met

abol

ic c

ost (

ml O

2 kg-

1 m-1

)

Speed (m s-1)

W

T G

4

2

3

1

0

�

Met

abol

ic c

ost (

J kg-

1 m-1

)

0 2 4 6 8 10 12

Metabolic cost of transportation (normalized for body mass) for different gaits of horsesMinetti, 1999

§ Each gait has a specific velocity that minimizes energy expenditure

§ Explains why animals and humans change gait for different speeds

Walking

Trotting Galloping

Modeling Legged Locomotion Dynamic vs. Static Locomotion

18||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Statically stable § Bodyweight supported by at least three legs§ Even if all joints ‘freeze’ instantaneously, the

robot will not fall§ Safe and slow

CoGCoG

Dynamically stable § The robot will fall if not continuously moving § Less than three legs can be in ground contact§ Fast and demanding for actuation and control

Modeling Legged Locomotion Dynamic Locomotion

19||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Modeling Legged Locomotion Dynamic Locomotion: Leg Structure

20||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Leg during running is not an inverted pendulum

§ Spring loaded inverted pendulum (SLIP)

Static locomotion (IP) Dynamic Locomotion (SLIP)

Modeling Legged Locomotion Dynamic Locomotion: Leg Structure

20||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Leg during running is not an inverted pendulum

§ Spring loaded inverted pendulum (SLIP)

§ are robust against collisions § can better handle uncertainties§ can temporarily store energy§ reduce peak power

[Alexander 1988, 1990, 2002, 2003]McGuigan & Wilson, 2003 – J. Exp. Bio.

Dynamic Locomotion SLIP Principles in Robotics

21||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Early Raibert hoppers (MIT Leg Lab), 1983§ Pneumatic piston§ Hydraulic leg “angle” orientation

Dynamic Locomotion From Raibert Hopper to Humanoids and Quadrupeds

22||Péter Fankhauser 27.02.2017Autonomous Systems Lab

§ Boston Dynamics§ Founded 1992§ Big Dog V1 (2005)§ Big Dog V2 (2008)

Dynamic Locomotion From Raibert Hopper to Humanoids and Quadrupeds

23||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Dynamic Locomotion From Raibert Hopper to Humanoids and Quadrupeds

24||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Series Elastic Actuators From Position to Force Controlled Systems

25||Péter Fankhauser 27.02.2017Autonomous Systems Lab

motor gear

position

link

spring linkmotor gear

position force

§ Kinematic, position control

§ Dynamic, force control

Series Elastic Actuators Efficient Walking and Running

26||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Series Elastic Actuators StarlETH – A Complaint Quadrupedal Robot

27||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Hutter, 2012 & Gehring, 2013

Series Elastic Actuators ANYdrive – A Robust and Force Controllable Robot Joint

28||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Absolute position sensing

Fully integrated

Programmablecontroller

High load bearing & hollow shaft

Ingress-protection IP66

Accurate position &torque control

ANYmal Combining Dynamic Motion Skills with Large Mobility

29||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Outdoor operation Water-proof & ruggedized design

Extreme mobility 360° rotation of all joints

Fast locomotion Dynamic gaits at 1 m/s

Modular payload 10 kg payload

Full autonomy Laser sensors for navigation

Long endurance 3 h operation

Safe and lightweight 30 kg and force-controlled

Hutter, 2016, Gehring, 2013, Fankhauser, 2014

ANYmal System Overview

30||Péter Fankhauser 27.02.2017Autonomous Systems Lab

Locomotion PC

12× ANYdrive IMU 4× Foot contact sensor

State estimation Legged odometry, ground est.

Locomotion Control Trotting, crawling, Free Gait

ROS interface

Inspection PC

Zoom camera 2× MicrophoneVisible & IR lightThermal camera

Visual inspection Gauges, levers

Thermal inspection

Audio inspection Pump & alarm sounds

Navigation PC

2× Laser rangesensor

2× Wide-anglecamera

Localization ICP, Rovio, Tango etc.

Terrain mapping Elevation mapping, traversability est.

Motion planning Navigation, foothold selection

Mission execution

Ethe

rnet

Wireless Router

Safety operator

WiFi

Radio

Operator PC

IndustrialWireless

Remote control UI Manual, supervised control

Visualizations Sensors, robot state, environment

Mission Mission creating and protocol

RT

Fankhauser, 2016

https://www.youtube.com/watch?v=SR5OJ-vklIs

Samuel BachmannDario BellicosoMichael BloeschRemo DiethelmChristian GehringMark HoepflingerGabriel HottigerMarco HutterJemin HwangboLinus Isler

Péter Fankhauserpfankhauser@ethz.ch

Thank You

32||Péter Fankhauser 27.02.2017

Dominic JudRalf KaestnerBruno KaufmannPhilipp KrüsiAndreas LauberPhilipp LeemannKonrad MeyerRoland SiegwartVassilios TsounisMartin Wermelinger

github.com/ethz-aslgithub.com/leggedrobotics

Open-Source Software

Autonomous Systems Labwww.rsl.ethz.ch www.asl.ethz.ch

||Autonomous Systems Lab 28/Feb/2017 1

Introduction to the V-REP robot simulator

||Autonomous Systems Lab

§ Introduction of V-REP scene§ Model browser§ Scene hierarchy (LUA child script)

§ MATLAB remote APIs§ Obtaining robot pose and sensor

measurements and sending control inputs.

§ Material can be found from http://www.asl.ethz.ch/education/lectures/autonomous_mobile_robots/spring-2017.html

§ You can build your own algorithms based on this.

2

Exercise #1 (14:15-16:00, 28/Feb/2017) in CAB G11

28/Feb/2017

V-REP scene

Remote API