Dlexa the Hedge Dragon
Dlexa HomeDlexa's BodyDlexa's MindFantasy Story for DlexaArtificial Inteligence is MathBuild Notes and Logs

Anatomy of the Hedge Dragon Body

Body Topics

The Escallonia hedge provides a 3D form for the dragon and the networked microcontrollers provide the nervous system for her sensors.

On this page we dissect the body of Dlexa into her flesh, sensors, activators, and the electronic nervous system that ties it all together. Her body is out in the weather on the south east corner of Dean's yard. Her mind (see that tab) is safe inside the house and safe from the elements.

Being out in the weather and within easy access of pets, wildlife and curious human hands means the components must be durable, easy to repair, and cheap to replace. If you spend many hours designing and building a component you must think about how often it will fail or be damaged.

You must build enough spare parts to span the expected survival period. You must make it easy for you to replace but hard for a "threat" to remove or reuse.

For example Dlexa needs eyes. You could buy a couple of $600 cameras but even in a nice city like Colwood they would probably disappear.

Dlexa uses $5 ESP32 cameras that come with a eFuse security key. Setting that key means the camera will only work with applications running on Dlexa's Mind. For physical security: casting the camera in a block of plastic makes them little more than paperweights. Proper design makes them easy to assemble and cheap to replace if lost.


The hedge is the cover that hides wiring and instrument groups so the sculptured shape and the wiring plan need to be coordinated to provide repair access.


Cameras provide both proximity detection and visitor recognition. Dlexa has two for binocular vision and these are part of the green-yellow cosmetic eyes (see the top banner of this page). For Pearl also binocular but completely hidden. For Worms only a single camera is required to detect a visitor.


Need lots of microphones to feed the STT application and provide noise cancellation. Dlexa will start with a microphone in each ear which may be augmented with a second pointing away from the visitor. If noise is a problem: these may be replaced by four microphone arrays with active noise cancelation at the array. Pearl will start with a pair in a sound baffle with one for the visitor and one sampling background noise (might even try a repurposing a sound canceling headphone mic. Worms will have several microphones to provide even sampling along the length of the Worm's home. All will be for visitors and the home "movie set" will be used as a baffle.


DML (Distributed Mode Loudspeakers) will provide the voice. The speaker for Dlexa's voice is the largest and built as a 2 foot by 2 foot plate (4 square feet) buried in her chest with two small 3 watt exciters. Pearl will use a 1.5 foot by 2 foot plate (3 square feet) in the shape of a gong hanging behind the Pearl tripod with one exciter. Worms will have a 6 inch by 4 foot wall (2 square feet) behind their home with two exciters at each end. These sizes are limited by the fact the plates must be suspended so an installation has to include gaps on all sides of the plate.

With a DML "larger" means better low frequency response. The active element (Corning Formular) is cheap and easy to cut so there is an incentive to find space for a largest plate possible. Poor low frequency response in the DML plate can be compensated by a bass-boost filter and more exciters but more square feet is always more efficient than more exciters.

Lights and Power:

All the lighting will be WS2811 driven strings or WS2812 buttons. A sheet of color can be a 5V or 12V strip of "dumb" LEDs driven by a single WS2811 with a FET current driver (many amps - many LEDs on a long strip).

Smaller strips (like inside the Pear ball) can use pieces of dumb strips with a ULN2003 driven by a WS2811 chip. The ULN2003 can drive 500ma but that is shared across the 7 channels. Setting a 5V strip all "white" (R+G+B) means using only 3 channels and limits to 160 milliamps or a strip 8 LEDs long. There are higher current drivers: TBD62083 is 500ma per channel but that chip has a 1 watt limit (so only 10 LEDs). Using a 12 volt strip means a 3 LED series only needs 20 milliamps so 24 LEDs from a ULN2003. Inventing an appealing design is of little value if you cannot power it.

Of course with the WS2812 strip each LED button has a tiny WS2811 inside so all you need is power and data and the LED strip can be 1000 LEDs long. A longer strip means longer time to update all 1000 LEDs. 33 FPS (Frames Per Seconds) is as fast as you can update 1000 LEDs and some people will notice that as flicker.

Power consumption is not a trivial problem. Each color (R, G or B) only needs 20ma. An RGB cluster (24 million colors) needs 60ma. A maximum strip with 1000 RGBs would need 60 amps. So the problem with lots of LEDs is not the data line it's the 5V power line. The 1.0 version of Dlexa will use an 8 wire cable providing 24 volts at > 2 amps (48 watts) on each pair. With multiple buck converters we can turn each 24 volts pair into roughly 10 amps at 5V. That is 40 amps at 5V for the cable.

We'll need max 2 amps to run all the electronics. The DML exciters need another 2 amps. Speakers are peak power things (2% duty cycle). If they are consuming 10 watts continuous the neighbours will complain. This means we can put 4 amps into lighting and that is 70 LEDs (if they are turned on hard (all 100% white). Typically a very bright colorful display is about 30% load so we can plan for 250 LEDs. A rough guess is Dlexa+Pearl+Worms on version 1.0 will be just over 100 LEDs.

40A would run all the electronics and speakers and leave enough for nearly 1000 LEDs all full 100% white. That many LEDs would be too much. Dlexa would look like a Las Vegas Casino and detract from her primary AI function.


The only movement planed is Dlexa's head. Her head will by grown in a long narrow pot. The head and pot will sit on a plate supported by a stand at the front of hedge. The plate is motorized so it can turn about 270 degrees. The pot is also mounted on a kind of seesaw that is supported by the plate. The seesaw can move the nose up and down about 45 degrees. This provides two degrees of freedom and the nose can be pointed at a visitor as they pass by the hedge.

The programming to have a visitor kept in the center of her view based on the binocular vision of her eyes is relatively simple. The rotation is just one motor on the edge of a lazy susan. The seesaw uses a center pivot and a chain drive pulls at front and back of the seesaw to position the nose. The motors are steppers so they can count the movement in steps. Out-of-range switches reset the zero point of the travel (up/down left/right). In the future TOF (Time of Flight) sensors (like those in electronic tape measures) could verify the position without have to force the head to an extreme (upper right) to zero the steppers on startup.


The Topic Links on the left open up to the detailed plans (technical drawings and calculations) and document the performance of the body parts. They will also document design changes when inventions didn't work.

Invention is really a form of art. Most of the time things do not work. Microphones are too noisy, motors are under powered, stuff takes up more room than expected, parts are too complex and as a result are not reliable or repairable. Who knew bunnies liked the taste of the chosen data cable? As Dlexa progresses those links will show her growing pains.