My research is on self-reproducing systems (SRP). Specifically I am trying to build
a robot / 3D printer combination that can (a) make two copies of itself and then
(b) forage for materials that can be used by those copies to make two copies.
The idea is that this leads to exponential growth. The goal is to terraform Mars.
Exponential growth is the only way to solve planet-scale problems.
A sub-problem is creating these SRP systems with software that learns.
By "learning" *I* mean self-modifying programs . Learning, in this system, is
defined as a permanent change based on experienced events. It is not the
current "machine learning" / "deep learning" model. Post-learning, the software
is running self-changed code. (This raises an interesting philosophical
question I'll address below ).
To make self-modifying programs I need a language that can self-represent.
Lisp is ideal for this. Because I am using such limited hardware I need to
make a space-efficient implementation. Forth is ideal for this.
To make systems that are efficient I need hardware that can self-adapt.
Field Programmable Gate Arrays (FPGAs) are ideal for this. I need an
adaptable CPU instruction set. I am using RISC-V as it is designed to
support extensions. In particular, I'm looking at extended instructions to
natively support Forth / Lisp.
There are a raft of problems that need to be solved. For example, the
"wire problem" or the "fastener problem". One question is how to
construct a servo motor. I'm looking at 3D printed metal  (or, more recently
3D printed glass  since quartz is easier to find).
Another research question is how to find material. It turns out that certain
plants can extract things like metals . Some of these metals are radioactive
which could provide a power source. An alternative idea is to develop
insects, such as ants, that prefer to build nests in various kinds of metal
rich soils, bringing the spoil to the surface. Quartz seems to be readily
available, making glass structures interesting.
The current state of implementation of the research is primitive so far.
o I have 2 robots and 2 3D printers. One of each will be disassembled to
provide models for 3D parts. This will obviously fail various criteria but
it provides a starting point for raising questions. I have to get better at
o I have experimented with 3D printed Japanese joinery parts. I have
to prototype constructing a multi-segment "joinery'd" arm.
o I have an FPGA I have programmed in Verilog. It runs RISC-V.
It will soon run Forth. The game is to get the FPGA to compute
and output an FPGA bitstream that can be re-loaded into the FPGA,
thus demonstrating self-modifying hardware. This also requires
careful thought about self-representation which is keeping me up
o I have to write FPGA routines to make an FPGA-ROS node capable of
controlling a joint. Each joint will contain a custom FPGA for control.
This will allow for joint adaptation. This is queued behind the prior problem.
o I have micro- and nano- controller boards running Forth. I have a 6
breadboard pseudo-joint layout (servos controlled by the microcontrollers).
I'm programming each joint to be a ROS node. I have to write ROS
node software in Forth. I also have to write Forth code to control the
Wifi / Bluetooth hardware. This will demonstrate non-wire joint control.
o I have Freecad and Gazebo software that can talk to ROS
so I can model the robot and use them to drive the pseudo-joints over
Wifi / Bluetooth. I have to prototype this setup. This will enable
merging real parts with simulated parts as the system evolves.
I have both Freecad and Gazebo running but need to demonstrate
remote ROS / Gazebo integrated control. This is the "top of stack"
o I have composite conductive PLA for the 3D printer [2a]
. I need to prototype embedding these "printed wires" into a structure.
I need to measure the voltage / current limits. I need to find / experiment
with other materials.
o I have been looking at 3D printed harmonic drives  to enable higher
torque from minimal motors.
o I have cameras and an iRobot Create3  platform. I have mounted a
camera and a Raspberry Pi 4. I want to experiment merging OpenCV
with the Freecad / Gazebo / Pseudo-arm over ROS. I need to prototype
the ROS / Wireless connection on the Pi. This will prototype the
post-reproduce material search behavior.
o I have a novel idea for end-point measurement of the robot hand.
I have linear diffraction gratings  and various lasers. This can project
spots that can be used to remotely measure the robot location. This
needs a spot sensor and needs to be prototyped.
o I have a hydroponic setup to grow sunflowers. I need to figure out
how to measure the uptake of metals I put into the soil. (I also have
free-range ants but they are not yet subject to test :-) )
It is all rather disconnected at the moment. However, it is all really just
a Gedanken platform for my omphaloskepsis  :-)
 If a system learns by self-modification it will be fundamentally
different from all other "copies". That means that it can't "teach"
other systems by simple copying since they will also have diverged
from the original. Thus it forces "learning by teaching" so that each
system can adapt "what it learns" based on "what it knows". Or to
put it simply: "You can't teach an old dog new tricks" because the
"old dog" is fundamentally different than it was.
 Japanese joinery
 The Wire Problem
We don't want wires if we can help it. Each joint requires power and signals.
Bluetooth at each joint can eliminate signal wires. The Proto-pasta company
makes composite conductive PLA material for 3D printers. The game is to
embed this conductive material in the physical structure of the robot to carry
power / ground without wires.
 The Fastener problem
We don't want fasteners since that requires tools. So the game is to use
Japanese joinery  techniques. I am prototyping 3D printed versions to
build the robot structure.
 Heavy Metals Extraction Potential Of Sunflower And Canola
If we can seed the planet with phytoextractive plants then we can use
the plants as a "mining operation". It appears that these plants can even
extract radioactive elements which will make a useful power source.
 Glass 3D Printing
Glass has interesting advantages. In particular, it is easily shaped and
can be made to fit areas without fasteners.
 TIRES - A Robot / Human Cooperative Effort To Change Tires.
This is research I did at CMU with Scott Fahlman.
 3D printed metal
There are various parts that need metal. In particular we need to be
able to create 3D printed servo motors.
 iRobot Create3
Foraging is a whole other problem that is still "on my horizon". But I
have a deep background in vision so it is easy to create a prototype.
 Navel Gazing
 Harmonic drives
 Linear diffraction grating
Lasers can be used with these gratings to patterns based on
position and orientation. This makes it ideal for remotely measuring
things like robot end-effector pose.