Automatic Design and Manufacture of Robotic Life CPSC 607 Yuan Luo - PowerPoint PPT Presentation

Automatic Design and Manufacture of Robotic Life CPSC 607 Yuan Luo

Outline  Background  Motivation  Golem Project: Methodology & Experiments  Results  Conclusion  Future Work

Reference: (2) GOLEM Project Website http:// demo.cs.brandeis.edu/golem (3) H. Lipson and J. B. Pollack (2000), "Automatic design and Manufacture of Robotic Life forms", Nature 406, pp. 974-978. (4) R. Brooks (2000) From Robot Dreams to Reality, Nature 406, pp. 946-947 (5) N. Forbes (2000) "Life as it could be: Alife attempts to simulate evolution", IEEE Intelligent Systems, December 2000, pp. 1-6

Background Alife vs. life Artificial life explores natural life by attempting to create biological phenomena in computers and other nonliving media. Common theme: explore the fundamental principles of life by building detailed working models. From robot dreams to reality One goal is the construction of living systems out of non-living parts. Carbon-chain based à Silicon-based, germanium- based,…

Motivation Most Alife scientists feel that the artificial simulation is the right way to go. Examined on the basis of understanding the principles, and simulating the mechanisms, of real biological forms. Aeroplane ßà Bird: share the same principles, but not the same implementation in chemistry

Motivation Biological life is in control of its own means of reproduction: auto-catalyzing chemical reactions But this autonomy of design and manufacture has not yet been realized artificially. Robots are still laboriously designed and constructed. Golem Project’s Goal: to realize artificial life, full autonomy attained not only at the level of power and behavior, but also at the levels of design and fabrication. Then, their own evolution, reproduction.

Golem Highlights Golem (Genetically Organized Lifelike Electro Mechanics) Project:  Conduct a set of experiments in which simple electro-mechanical systems evolved from scratch to yield physical locomoting machines.  Take advantage of the nature of their own medium: thermoplastic, motors, and artificial neurons  Use evolutionary computation for design, additive fabrication for reproduction  This is the first time robots have been designed and robotically fabricated

Evolutionary Design The evolutionary process operates on a population of candidate robots, each composed of some repertoire of building blocks. The evolutionary process iteratively selects fitter machines, creates offspring by adding, modifying and removing building blocks using a set of operators, and replaces them into the population.

Evolutionary Design Building Blocks: bars and actuators Building Blocks of Control: artificial neurons Bars connected with free Joints form trusses Neurons can connect to create arbitrary control architectures, Bars’ number, the way of connecting, neurons’ number, synapse’s weights, activation threshold, etc.. can be changed by mutational operators

Evolution Simulation • Robot representation • Fitness function: The net Euclidean distance that the center-of-mass of an individual moves over 12 of cycles of its neural control

Evolution Process Simulation • Starting with a population of 200 null (empty) individuals. • Each experiment used a different random seed. Individuals were then selected, mutated, and replaced into the population. • selection function: fitness-proportionate selection replacement function: random replacement mutation operators: independently applied with following probabilities (length of bar or neuron synaptic weight(0.1), removal or addition of a bar(0.01), etc..) • The process continued for 300 to 600 generations A typical evolution process looks like : ./evolution.mpg

Simulator Both the mechanics and the neural control of a machine were simulated concurrently. The mechanics were simulated using quasi- static motion, where each frame of motion is assumed to be statically stable. The model consisted of ball-joined cylindrical bars with true diameters. Each frame was solved by relaxation: define an energy term, taking into account elasticity of the bars, potential gravitational energy, and penetration energy of collision and contact. The neural network was simulated in discrete cycles. In each cycle, actuator lengths were modified in small increments not larger than 1cm.

A sampled instance of an entire generation:

Evolution Simulation Various patterns of evolutionary dynamics emerged some of which are reminiscent of natural phylogenetic trees. (a) Extreme divergence: niching methods (b) Extreme convergence: fitness-proportionate (FP) selection (c) Intermediate level of divergence: early stage of FP selection (d) Massive extinction under FP selection

Automated fabrication Selected (virtual) robots out of those with winning performance were then automatically converted into physical objects. The manufacturing process uses commercial rapid- prototyping technology ("3D Printing"), which generates the entire structure layer by layer.

Automated fabrication Procedures: (1) The bodies, which exist only as points and lines, were first converted into a solid model with ball-joints and accommodations for linear motors according to the evolved design. (2) The virtual solid bodies were then materialized using commercial rapid prototyping technology. Plastic deposition 3D printing machines use a temperature-controlled head to extrude thermoplastic material layer by layer, so that the arbitrarily evolved morphology emerged as a solid three-dimensional structure without tooling or human intervention.

Some fabrication examples  Cylinder: • Full Robot:

(3) The entire pre-assembled machine was printed as a single unit, with fine plastic supports connecting between moving parts.

(4) The resulting structures contained complex joints that would be difficult to design or manufacture using traditional methods. (5) The evolved controller (neural net) is downloaded into a PIC microcontroller. Standard stepper motors are then snapped in, and the evolved neural network is executed on a microcontroller to activate the motors After the robot performs, it can be melted and recycled into another form for the next task .

Results In spite of the relatively simple task and environment surprisingly different and elaborate solutions were evolved.

Results Not less surprising was the fact that some exhibited symmetry, which was neither specified nor rewarded for any where in the code; a possible explanation is that symmetric machines are more likely to move in a straight line, consequently covering a greater net distance and acquiring more fitness. Similarly, successful designs appear to be robust in the sense that changes to bar lengths would not significantly hamper their mobility

Conclusion Simple from the perspective of what human teams of engineers can produce, and what biological evolution has produced BUT for the first time a robotic bootstrap, where automatically designed electromechanical systems have been manufactured robotically; minimized human intervention both in the design and in the fabrication stages.

Conclusion  Using a different medium and evolutionary design in simulation, they have made progress towards replicating autonomy of biological design and reproduction.  This is the first time any artificial evolution system has been connected to an automatic physical construction system.  All together, the evolutionary design system, solidification process, and rapid prototyping machine form a primitive "replicating" robot .

Future Work will this process progress beyond toys to more complex machines?

There are several approaches to the question of sustained evolution.  Very large populations and lots of CPU power, might still make progress. In particular, individuals with decoupled functionality might find it easier to adapt, thereby promoting modularity without explicit intervention.  It is possible that creatures reach some maximum level of complexity because the environment is too dull - in experiments only an infinite plane. Perhaps with inclusion of more sophisticated environment, more building blocks like sensors, and interaction between robots, might promote more complexity .  The design space is exponential - as more components are added, the permutations of parameters becomes so large that the space of possibilities becomes intractable. However, if we can find a way for the evolutionary process to discover and reuse modules, then the complexity of the building blocks increases exponentially too, and the design space might be tractable after all.

Questions!!!