Autonomous walking machine under neural control

[AMOS]

AMOS-WD06


Fig.1


Fig.2






The AMOS-WD06 (Advanced Mobility Sensor Driven-Walking Device) is the reactive 6-legged walking machine which mimics the structures of walking animals.
This walking machine driven by a modular neural controller (Only one artificial neural netwrok) can autonomously perform reactive behaviors, e.g. wandering around, avoiding obstacles, escaping from deadlock situations or corners, reflex action (standing in an upside-down position) and auditory- and wind evoked escape behavior (run away from a wind puff and/or an auditory signal). Furthermore, it can also walk in omnidirection [1] generated by the neural controller.

It can serve as a hardware platform for experiments concerning the function of a neural perception-action system [1,3].

Technical details:

This robot has now 28 sensors, 21 motors.

It consists of six identical legs. Each leg has three joints (three DOF): the thoraco-coxal (TC-) joint enables forward (+) and backward (−) movements, the coxa-trochanteral (CTr-) joint enables elevation (+) and depression (−) of the leg, and the femur-tibia (FTi-) joint enables extension (+) and flexion (−) of the tibia (see Figure 1). Each tibia segment has a spring-like compliant element to absorb impact force as well as to measure ground contact during walking. All leg joints are driven by analog servo motors. Inspired by invertebrate morphology of the American cockroach’s trunk and its motion, the trunk of the machine is formed with two body parts (see Figure 2) : a front part (T1) where two forelegs are installed and a central body part (T2) where two middle legs and two hind legs are attached. They are connected by one active backbone joint driven by a digital servo motor. This machine has six foot contact (FC1,...,6) sensors, seven infrared (IR1,...,7) sensors, two light dependent resistor (LDR1,2) sensors, one gyro (GR) sensor, one inclinometer (IM) sensor, one upside-down detector (UD) sensor, and one auditory-wind detector (AW) sensor . The foot contact sensors are for recording and analyzing the walking patterns [1]. The IR1,...,7 sensors are used to elicit negative tropism, e.g., obstacle avoidance and escape response [1], while the LDR1,2 sensors serve to activate positive tropism like phototaxis [2]. The GR and IM sensors apply to upward/downward slope detection. The UD sensor is employed to trigger a self-protective reflex behavior when the machine is turned into an upside-down position [1]. The AW sensor is applied for the auditory-wind application. It will activate the auditory- and wind-evoked escape behavior of the walking machine.

The control (artificial neural network [1,3]) of this walking machine is programmed into a mobile processor (a PDA). The PDA is interfaced with the MBoard which digitizes all sensory

signals and generates a pulse width modulation (PWM) signal to command all servomotors. The communication between the PDA and the MBoard* is accomplished via an RS232 interface at 57.6 kbits/s.

Electrical power supply of the whole systems is via batteries of one 7.4 V Lithium polymer 2200 mAh for all servomotors, two 9 V NiMH 180 mAh for the electronic board (MBoard) and a wireless camera, and four 1.2 V NiMH 2200 mAh for all sensors.

*MBoard is the Multi-Servo IO-Board (MBoard). It can control up to 32 servo motors and can digitized analog sensory signals up to 36 signals. Its size is 125 mm x 42 mm x 15 mm.

Programming:

• C programming on the MBoard for controlling servomotors and for reading digitized sensor data
• C/ C++ programming of the neural controller [1,2,3] created in Embedded Visual C++ for implementing on a PDA.

References:
[1] P. Manoonpong, F. Pasemann and F. Woergoetter, “Sensor-driven neural control for omnidirectional locomotion and versatile reactive behaviors of walking machines,” Robotics and Autonomous Systems, vol. 56, no. 3, pp. 265–288, 2008.
[2] P. Manoonpong, F. Pasemann and F. Woergoetter, “Reactive neural control for phototaxis and obstacle avoidance behavior of walking machines,” International Journal of Mechanical Systems Science and Engineering, vol. 1, no. 3, pp. 172–177, 2007.
[3] P. Manoonpong, “Neural preprocessing and control of reactive walking machines: Towards versatile artificial perceptionaction systems,” Cognitive Technologies, (BOOK), Springer, 2007.

Check the following links to see its performance:


http://www.nld.ds.mpg.de/~poramate/AMOSWD06.html


Omnidirectional walking:
http://www.nld.ds.mpg.de/~poramate/RAS/OmniR.mpg
Outdoor test:
http://www.nld.ds.mpg.de/~poramate/RAS/OutDoorTestRunResize.mpg


Obstacle avoidance behavior:
http://www.nld.ds.mpg.de/~poramate/R...eAvoidance.mpg

The same controller is also applied to 4 legs and 8 legs:
http://www.nld.ds.mpg.de/~poramate/BuiltWalkingMachine/AMOSWD06/NeuralPreprocessingAndControl_ArtificialPerception _Action%20Systems.avi

http://www.nld.ds.mpg.de/~poramate/RAS/OmniR.mpg

[ame="http://www.youtube.com/watch?v=xfyYfhJoTxk"]YouTube - Autonomous walking robot (AMOS-WD06) under neural control[/ame]
[ame="http://www.youtube.com/watch?v=g2bduFRgieo"]YouTube - Phototaxis: Autonomous walking under neural control[/ame]

[darkback2]

I checked out your website. Some of the behaviors you have programmed are really cool. I'm going to steal them for my hexapod as soon as its done! No...seriously. I especially like the ideas on escaping predators, and the inversion reflex.

Some Ideas if you don't mind:

- When my cats feel threatened they tend to crouch down lowering their center of gravity. My dogs do the same thing.

- I wasn't sure if it displayed a strafing gate as a method of avoiding obstacles. It seamed to turn away, and when it was finding its way out from behind the stairs it looked like it may have taken a few steps to the side, but I'm not sure.

- Also, How does it know where it is going, as in can you tell it to go somewhere? or does it wonder around randomly.

- Finally...this is a wonderful modeling platform for animal behavior. Have you thought about making it into a predator? Perhaps having it seek out smaller robots around the house, tip them over, and steal their batteries?

Just a thought.

All in all, this is an incredible project. I hope you do really well with it in this contest. You should also bring it to robogames 2009. It would do really well in some of the different categories.

DB

[4mem8]

Interesting robot AMOS well done.

[AMOS]

Hi, Thank a lot for all comments and nice idea. Well, in general AMOS just walks randomly and avoid obstacles but as soon as it detects a light source, it will turn toward and approach it (positive tropism). The idea is that this robot should not only autonomously perform various behavior inspired by animal's behavior but also show some adaptive behaviors (coming soon ).

[Adrenalynn]

What an excellent contest entry for this round! I watched the video of it escaping deadlock and corners - outstanding job!

[Zenta]

Hi,

Excellent work! I saw you made a 4 and 8 legged version too! Awesome!
Do you have a video of your AMOS walking over obstacles? I saw some pictures of it but didn't find a video.
Have you implemented the kinematics for the extra T1 - T2 backbone joint?
What kind of servos do you use?
What is the total weight?

Very great work!

[AMOS]

Hi, The total weight of it is about 4.2 kg. including all electronics, battery packs and PDA.
It can autonomously climb over obstacles having a hight of up to 5.5 cm by using its backbone joint controlled by also a neural network. By the way, I have not filmed it yet but soon : ) .

By the way, the interesting of this work is the neural controller where it can apply to 4, 8 ,and 6 legs without changing its internal structure.

see more movie at http://www.nld.ds.mpg.de/~poramate/BuiltWalkingMachine/AMOSWD06/NeuralPreprocessingAndControl_ArtificialPerception _Action%20Systems.avi

[4mem8]

Amazing work Amos, Thanks for posting your project.

[Zenta]

Hi, The total weight of it is about 4.2 kg. including all electronics, battery packs and PDA.

Wow! That's heavy! What servos did you use?

It can autonomously climb over obstacles having a hight of up to 5.5 cm by using its backbone joint controlled by also a neural network. By the way, I have not filmed it yet but soon : ) .

You have to film that, it would be really great! About the neural network, it handle the inverse kinematics too?

By the way, the interesting of this work is the neural controller where it can apply to 4, 8 ,and 6 legs without changing its internal structure.

Thats great! Very interesting. Have to study the neural part one time...
have you tried implementing other gait methods like the ripple gait (or scorpion gait for the octopod) instead of the tripod gait?

-Zenta

[AMOS]

//Wow! That's heavy! What servos did you use?

I use a servo from bluebird. Its weight is about 45 g but most heavy parts are from construction of a robot .

//You have to film that, it would be really great! About the neural network, it handle the inverse kinematics too?

I will film it soon. I don't use forwards/ inverse kinematic because such method will need a lot of computation power and (I am not sure) it may require a lot of memory comparing to the neural network model I used.

On the other hand, the neural network of AMOS performs as Central Pattern Generator requiring less computation power. It basically generates oscillating signals for driving motors.

//Thats great! Very interesting. Have to study the neural part one time...
//have you tried implementing other gait methods like the ripple gait (or scorpion gait for the octopod) //instead of the tripod gait?

Yes, AMOS can not only perform autonomous reactive behaviors as you have seen on videos but also walk with at least 5 different gaits including SLOW WAVE GAIT, FAST WAVE GAIT, TRANSITION GAIT, TETRAPOD GAIT, TRIPOD GAIT. All these versatile autonomous behaviors and a variety of gaits are achieved by one modular neural network.

Unfortunately, I cannot officially provide videos because I have submitted an article about this method to some journals where they ask to keep videos confidently until the article releases.