Mobile Breadboards for Nervous Network Sensor Integration

Mobile Breadboards for Nervous Network Sensor Integration

To market home robots successfully they have to be durable, efficient, and smart enough to live in your house without damaging your stuff. Designers tend to put so much into the internals of the robot we forget we have to live with them if they’re going to do work for us, and that means finding ways they can bumble about without pissing us off.

To this end I wind up building a lot of mobile breadboards to explore survival competence using various low-resolution commercial optic and touch sensors. One of my latest is the “Huey” line of experimental rovers for studying long term home survival. Hybrid powered (battery and solar), the device is designed to explore minimal configurations by optically guiding itself about a house, trying to avoid bumping into everything by following light sources and shadows.

Tilden-Huey01 - components
Initial components for the Huey autonomous rover.

Custom anodized aluminum parts and dedicated PCBs allow for reliable prototypes with decent de-acceleration trauma resilience (i.e., getting accidentally kicked).

The key component is the blue cell phone charger tube I picked up in a Chinese junk market for $3US. It takes one standard AA battery and pumps it to a regulated 5v, 700mA at about 80% efficiency. Not the best but an incredible weight savings over conventional battery clusters so I can keep the design light. Works from rechargeables as well. You might find similar tubes in many surplus markets but note they generally require re-work to enhance their contacts, circuits, and reliability. Not for the electronically squeamish, but they’re cheap, clean, tough, waterproof, and match my machine parts, so they’re a staple on many of my research prototypes now.

Tilden-Huey02 - rough assembly
Huey’s rough component assembly.

The upper hole will contain a hand-built 8-element radial retina (4 directions, peripheral and focal sensors) so it can see differential shadows at a 3 meter distance or broad 180 degree fields. One of the first things you learn in visioned robots is the higher the eye, the less ground you have to ignore. So this bot has vision that can selectively fill its brain with inhibitory or excitatory information to (hopefully) make clever path decisions.

Other sensors will include a forward-backward optical-difference detector, three touch-antennae, and a whip antennae to reduce motor power if it starts to topple over.

Parts seem to fit. Now it’s just a matter of wiring it.

Tilden-Huey03 - assembled nacelle
One of Huey’s motor nacelles with attached 5 watt shortless motor-driving multiplexer PCB.

There are a thousand ways to make motor drivers but this one has color coded input-outputs and left-right symmetry to make hookup intuitive and quick. My basic symbol for it is as follows:

Symbol for
Symbol for multiplexer element.

The motor used is a NASA surplus Micromo gearmotor – horribly over-rated for this application but with a thousand hour lifetime, modified with lithium-grease and 10W-30 oil making it quiet and efficient. As Huey will be in operation for quite a while the extra motor expense is worth it, especially during the week-long demonstrations in front of colleagues and sponsors it will undoubtably suffer.

And quiet robots? If you live with them, they better not be louder than a cat in operation, as one of my standard rules. :)

16 element
The 16-element Nervous network (Nv) brain board with set-reset switches.

Finding a decent batch of these surplus floppy-disk detection switches was a major motivator for this design. With these, it’s dead simple to configure one of the several thousand possible process-patterns from the Nv loops in a variety of architectures.

As background, Nv differentiating neurons are as follows:

The primal Nv neuron.
The primal Nv neuron.

And by arranging them into loops, different fundamental process patterns can be configured to drive motors in push-pull arrangements. This single chip Nv Hexcore can support 5 distinct patterns that coincidentally equate quite closely to general insect walking patterns.

A 74C14 hex Schmitt trigger chip.

The different process patterns exponentially explode after as few as 8 neurons.

Nv neural patterns.

But as the RC timing value of each neuron can be set individually by bias or sensor input, the actual number of adaptive patterns is very large. So in 16 neurons, does little Huey have enough brains to show off interesting behaviors? Yes, yes it does.

Tilden-Huey07 - parts assembly
Systems integration testing on a black leather casting couch, for electrical isolation.

One of the things I like about symmetric, color coded Nv designs is you can re-wire them anywhere on the fly, just to try a new parameter or behavior. Being able to change and test a bot’s mind within your interest window is one of the best methods I’ve found for accidental discovery.

Whether constructively or destructively, casual messing with robot brains is fun, especially when they do things right, but especially when they do things right under duress.

The finished 15cm tall robot (30cm with antennae) sporting a sample Braitenberg-style follow-the-light-avoid-the-object configuration.
The finished 15cm tall robot (30cm with antennae) sporting a sample Braitenberg-style follow-the-light-avoid-the-object configuration.

Huey’s battery life turns out better than expected at 8-10 hours with a rechargeable, and 12 hours with an Alkaline AA. Of course if it makes it to a sunny spot, the Sunceram solar cell (on back) and 5 Farad supercap take over, meaning it will run all day until it gets trapped. In the smartest power configurations, the battery just acts as a “glucose store” to get it from one interest-source to another across energy wastelands. As sunbeams move slowly across floors and walls, Huey’s vision system allows him to find new paths around obstacles, only rarely hitting them.

Of course I don’t watch it all the time. I generally notice it as it wanders about the computer desk or marches below the screens while I watch TV (attracted by the infra-red I suspect). I take note on how easy it is to live with, and so far, it meets expectations.

An active running experiment showing process patterns in the dark. The green eye LED allows me to find it when it gets trapped under chairs or tables.
An active running experiment showing process patterns in the dark. The green eye LED allows me to find it when it gets trapped under chairs or tables.

I make my custom Nervous Network boards with multiple surface mount LEDs, generally called “BLIFNARs” (Blinking Lights For No Apparent Reason), an acronym from my special-effects days. These should be more accurately called “BLIFARs” as they do serve a purpose of letting me know what’s going on in the creature’s brain as it wanders about or as I’m wiring it up.

Yes, analog allows you to work on live brains – a feature. Problem is that even after years of study, correlating complex behaviors with LEDs is difficult as so much goes on at the same time (sensors, processes, iterative pivot-points), especially when it’s wandering, especially when it’s struggling, especially when you can’t help it as that would defeat the purpose.

One of the biggest challenges when researching autonomous mind children is to avoid interference, testing how well they can stand on their own two servos, as it were. Gotta let them make their own mistakes; then you observe; then you build new machines with better reflexes and automatic competences. That’s how you proctor a minimalist species into existence.

(Thank you, David Brin.)

Tilden-Huey10 - Wandering around
Huey 1.2 on a wander patrol, the whip antennae acts as a balance sensor to keep it upright over rough surfaces by reducing motor power.

One of the advantages of an over-tall vertical walking breadboard is that if it gets into real bother, it just falls on its back like a turtle (like many pet owners, I’ve learned that no noise is generally a sign of trouble). This feature keeps this particular long-active robot from causing major wear damage to furnishings and carpet.

And itself, naturally, but that’s easy to repair. My carpeting isn’t. I know from experience.

For more information on BEAM Robotics and Biomech research in general, please Google “Mark Tilden Robot” or look up “BEAM Robotics” and check out some of our 25 year history, plans, kits, science, or millions of entertaining humanoid robots available now in your grocer’s freezer.

Or online shops. That works too.


This week marks the official launch of Make: Volume 39 — Robotics, which drops on newsstands the 27th. Be sure to grab a copy at a retailer near you, or subscribe online right now and never miss another issue.

We are celebrating with five days of robot-related articles, pictures, videos, reviews and projects. Tune into this space for Robot Week!

Our next theme week will be wearable electronics. Send us your tips or contributions before it gets here by dropping a line to


7 thoughts on “Mobile Breadboards for Nervous Network Sensor Integration

  1. Mobile Breadboards for Nervous Network Sensor Integration | Salute says:

    […] Read more on MAKE […]

  2. Making Friends – A BEAM Robot Pet | MAKE says:

    […] with a minimalist aesthetic can often result in complex analog behavior. Keep an eye open for a post about Huey, one of Mark’s latest robots, later […]

  3. richfiles says:

    Gotta love the design on this one! I put socket pins on the mux of one of my robots as well to experiment with gait patterns. One thing I have been looking into is using more complex networks to perform more complex functions. My solution, is not to have a bunch of jumper wires, but to feed the outputs of ever 74HC14 to the input of a Xilinx CPLD chip, and then feed the outputs to the RC network and input of every inverter, I can then connect input to outputs on a PC and program the chip with the nervous/neural network map. It is BEYOND overkill for what the chip can do, but it is in essence, the most advanced form of MUX chip you could ask for… and since the chip DOES feature logic capability, your network map can change, on the fly, based on sensory, command, or memory inputs!

    I had also come up with a design that used a pair of 5 bit accumulators to create a weighted neuron that fires with increased ease as it fires more frequently, and loses the ability to fire as easily when not fired frequently. It was entirely digital, and relied on a “decay pulse rate” being fed into it to set it’s “forgetfulness”, but I had the circuit compact enough to fit 6 of them on the smallest Xilinx CPLD chip. With 6 of them, you could have a pair of 74HC14s as your integrators to take incoming pulses and convert them into an overall excitatory or inhibitory pulse (without integration into the synapse like Nv/Nu style neuron, straight up pulses ORed into the chip via diodes or resistors could overlap, and not be taken as two pulses.

    I need to re-record the design, as I lost my papers on it.

    Of course… If we could get memristors… That’d be even better! That’s another project I’m working on. Point contact style memristors that can be made in a home lab.

    1. Tom Blitch says:

      Sounds awesome man, I remember your old website where you used Xilinx chips to make a spyder type walker. Ive been looking for excuses lately to use them myself. Do you regularly update your site anymore?

      1. richfiles says:

        I have not actually updated content at my site for quite a while, though I VERY MUCH wish to do so. Three jobs will do that to you…

        I have been progressing with my memristor idea. Did a second design. I thought there was potential for it, but I have to consider the pressure put on the point contact junctions. I had ideas involving sandwiched PC boards, aluminum ball bearings, and copper traces with a sulfide junction layer. Problem is they are super sensitive to over pressure.

        I’m now thinking a very thin PCB (0.5mm???) with slots cut to create star patterned “fingers” may provide suitable springs that would eliminate the issues of un evenness disrupting and shorting out junctions with the application of too much static pressure between boards as the two are tightened against one another. At the moment, my design has no springiness to it. This, I think solves the issue of springs. Problem is simply finding a board manufacturer that can cut such complex internal shapes. Maybe a place with laser cutting?

        Interestingly… I JUST got that idea right now, having replied to this comment of yours, and seeing this article again, thinking of the hexagonal patterns often used in Tilden’s PC board designs!


        I better go to bed…

        I have a lot of work to do tomorrow! :)

  4. Robot Week Wrap-Up | MAKE says:

    […] Mark W. Tilden […]

  5. H U N T E R says:

    I want to read more from Mark Tilden!

Comments are closed.

Discuss this article with the rest of the community on our Discord server!
Mark W. Tilden

Mark W. Tilden is perhaps best known as the inventor of BEAM robotics and the WowWee Robosapien humanoid robot. He is a robotics physicist who produces complex robotic movements from simple analog logic circuits, often with discrete electronic components, and usually without a microprocessor.

View more articles by Mark W. Tilden


Maker Faire Bay Area 2023 - Mare Island, CA

Escape to an island of imagination + innovation as Maker Faire Bay Area returns for its 15th iteration!

Buy Tickets today! SAVE 15% and lock-in your preferred date(s).