This is a video of my Futaba RBT-1 Robot performing a demo routine.

In the video you can see that I am controlling the body position of the robot using an accelerometer. The accelerometer will soon be attached to the robot and ultimately help with balance calculations.

This is the new body shell for my quadruped robot. I’ve also added some wiring sleeves to tidy up the cabling. One of the legs is dragging a little as the screws had come slightly loose. But now the robot is pretty much built I can finally start to concentrate on getting the walking gait smoothed up. I also need to think about what type of sensors to add for 360 degree object awareness.

This video shows Q-P2 running a Creep Gait routine.

I think I need some kind of rubber tip on the end of the legs as you can see him slipping on the wooden floor. I might even add some pressure sensors to the tips so I can work out feet position over rough terrain … hmmmm …

This Creep Gait is basically modeled on a cat’s creeping motion. That is to say at any given time one leg is lifted and moving forward while the remaining three are gradually moving back. The problem with getting a four legged robot to walk smoothly is maintaining a stable tripod with the three legs on the ground. To compensate for any loss of balance the center of gravity can be shifted to sit inside the tripod by moving the body side to side. I have not added any balancing at this stage as the position of the trailing feet just about encapsulate the center of gravity. However, I will probably add a bit of swing in the near future to make the creeping even more stable, especially if I want to get the robot to walk over an uneven surface.

Just a quick video demonstrating my robot doing some basic ‘press-ups’ using an IK routine. Finally got my servos sorted -- using 8xHS-645MG and 4xHS-485HB. Next I have to start working on a smooth creep gait sequence …

The chassis is built from Rigid PVC foamboard (thanks for the tip Innerbreed). This material is amazing for prototyping – it took me a whole two hours to design and cut out the two chassis components and the four end leg pieces. However, this is still a prototype and I when I’m happy with the design I want to get some black anodized aluminium parts laser cut instead.

I opted to use a Dagu ‘Red-Back’ Spider controller as I am already familiar with the Arduino and Processing environment. The board is based on an ATmega1280 MCU and can drive up to 48 servos. It’s nice to have everything all on one board with loads of room left for adding sensors etc.

Again – I used LynxMotion black anodized aluminium servo brackets to create the leg sections. There are 8 x Hitec HS-645 MG servo’s for the Coxa and Femur joints and 4 x crappy TowerPro 995MG for the Tibia servo’s. But as soon as my robot tin is full I will replace the TowerPro with another 4 x HS-645 – as they are the ultimate!

The whole concept and design is ‘heavily’ based around jezek68′s quadruped robot: http://www.youtube.com/watch?v=jWP3RnYa_tw. And if I can get my QP-2 to walk anywhere nearly as gracefully as you can see in that video then I’ll be a very happy man. I’ve also been studying gait generation algorithms and lots of AI and mathematics over the last year so hope to have some pretty cool things to show over the coming months.

After waiting for what seemed like an eternity my servos finally arrived in the post!! This has enabled me to put my Quad back together and start working on the gait programming. Here is a quick video to show what I’ve done so far.

Creep Gait

So in the video you can see the robot creeping forward. This type of gait is pretty much the same as what your cat does when it’s creeping up on something … slowly moving one leg at a time while making a stable tripod with the other three legs. I found Oricom Technologies NICO project a really useful resource whilst researching suitable walking patterns for my quad. There is a ton of really useful stuff but most of all the timing diagrams came in really handy. In order to implement the timing I wrote some code to produce trapezoidal curves over the course of a set period of ticks to produce the ‘y’ value for leg lift and ‘z’ value for travel:

        while (tick < period) {
            if (tick >= this.a && tick < this.b) {
                double AX = this.b - this.a;
                double BX = this.amplitude;
                double AB = Math.sqrt((AX * AX) + (BX * BX));
                double angle = Math.asin(BX / AB);
                y = Math.tan(angle) * (tick - a);
            } else if (tick >= this.b && tick < this.c) {
                y = this.amplitude;
            } else if (tick >= this.c && tick < this.d) {
                double CX = this.amplitude;
                double DX = this.c - this.d;
                double CD = Math.sqrt((CX * CX) + (DX * DX));
                double angle = Math.asin( CX / CD);
                y = Math.tan(angle) * (d - tick);
            } else {
                y = 0;
            }
            yield(y);
            tick++;
        }

... where 'a' is the start of the rise, 'b' is the maximum, 'c' is the fall of the rise and 'd' is the end of the rise.

I then used Inverse Kinematics to position the feet where they fall on the timing diagram. I based the IK on the algorithm described on the Learn About Robots website - but had to modify it so it would work in 3D. My next task will be to incorporate the body physics into the IK routine which will allow for tilt and twist (as per the original Phoenix hexapod code). This will be essential as I plan to invest in a gyro/accelerometer for balance compensation at some stage in the future.

Light & Sound

I'm using a small piezo speaker controlled by my arduino for the amazing sound effects! So far there is a few beeps when the robot finishes the setup routine and a repeated beep when it encounters an obstacle. I was also sent a free RGB LED from the good folk at oomlout.co.uk which is responsible for making the robots colourful expressions.

Sensors

The only sensor I have at the moment is a Sharp Infra Red detector. This generates a voltage representing the proximity to an obstacle within a range of about 10-30 cm. I'm not doing anything exciting when an obstacle is encountered at the moment aside from stopping the robot in his tracks and flashing the LED purple. I could perhaps sweep the sensor left to right and decide which way to turn to avoid the object - but I think I plan to use sonar for general obstacle avoidance instead. The IR sensor could then be used to detect any sudden drops in front of me or if the robot has somehow left the floor (ie - stairs or picking it up).

While waiting to save up for my new servos, I decided to put together this “BRAT” biped with parts from my Quad. Here is a video of his first steps …

You can probably see that I had to once again revert to using my trusty meccano for several parts -- namely the feet and the torso. I used two HITEC HS-645MG and one Tower Pro MG995 servos for each of his legs. The SSC32 was used as the controller which is tethered to my laptop to receive servo commands.

There probably won’t be much of a future for this little bot as I’ve just dismantled him so I can recover the bits for my Quad. However, if I ever come across any spare brackets and servos I think I’ll try using my Arduino to send a sequence of moves to the SSC32 instead of using the raw SSC32 sequencer to control him. After that would naturally come the sensors, AI and automation

For now, I hope you enjoy the funky video and the pictures in the gallery below …

The small body footprint of the initial design was causing me problems -- namely the coxa joints were too close together limiting servo movement. I’ve decided to use a piece of trusty meccano to serve as a new body until I can afford to get a custom aluminium body cut for me.

With a larger body I now have room to mount the electronics more effectively and there is also spare room for adding funky stuff such as sensors, head etc in the future.

There is also enough room to mount the battery pack beneath the body using a thick cable tie. This allows me to shift the battery pack forward and back allowing fine adjustment to the center of gravity.

I’ve managed to hack an old fat XBOX controller to use as my new control mechanism. This is a temporary solution until I can afford a blue-tooth receiver. I’ve done this by re-wiring the controller -- using the six wires in the default cable -- one for each of the analog pot data lines (LS horz/vert, RS horz/vert) the other two for power and ground.

Unfortunately, 2 of the 12 cheap Tower-Pro servos have burnt out -- limiting what I can do in terms of development of the gait and body movement. I’ve decided to invest in the superior quality Hi-Tec HS 645MG heavy duty metal gear servos. However, these come at a price so will probably purchase 4 at a time over the next year.

This is a quick video to show progress on my Quadruped Q-P1. My SES chassis arrived today Spent a few hours figuring out the best configuration and went ahead and built it up. Just added the servos and electronics to the chassis and this video demostrates the first power on test with some up and down movement and a drunken walk …

As you can see there’s still lots to do -- but I’m happy that it’s most definitely a robot now!

To do:

• Calibrate software with measurements and angles.
• Develop a better control system.
• Implement the rest of the body movement routines -- pitch, roll etc.
• Tidy up robot -- wires, electronics etc.
• Try to figure out a smooth walking gate -- a slow creep would be nice.
• Add some sensors
• Go autonomous!!!

I will post more when I accomplish some of the things on the list -- together with more comprehensive notes and hopefully some sourcecode to share. Not sure when I’m going to find the time over the next few weeks to do much of this as I’m wrapping up ready for a big move down south. But any comments are most welcome -- just use the contact link and drop me a line …