Jekanthan Thangavelautham

A coarse-coding framework for a gene-regulatory-based artificial neural tissue

A developmental Artificial Neural Tissue (ANT) architecture inspired by the mammalian visual cortex is presented. It is shown that with the effective use of gene regulation that large phenotypes in the form of Artificial Neural Tissues do not necessarily pose an impediment to evolution. ANT includes a Gene Regulatory Network that controls cell growth/death and activation/inhibition of the tissue based on a coarse-coding framework. This scalable architecture can facilitate emergent (self-organized) task decomposition and require limited task specific information compared with fixed topologies. Only a global fitness function (without biasing a particular task decomposition strategy) is specified and self-organized task decomposition is achieved through a process of gene regulation, competitive coevolution, cooperation and specialization.

A Neuroevolutionary Approach to Emergent Task Decomposition

A scalable architecture to facilitate emergent (self-organized) task decomposition using neural networks and evolutionary algorithms is presented. Various control system architectures are compared for a collective robotics (3 × 3 tiling pattern formation) task where emergent behaviours and effective task -decomposition techniques are necessary to solve the task. We show that bigger, more modular network architectures that exploit emergent task decomposition strategies can evolve faster and outperform comparably smaller non emergent neural networks for this task. Much like biological nervous systems, larger Emergent Task Decomposition Networks appear to evolve faster than comparable smaller networks. Unlike reinforcement learning techniques, only a global fitness function is specified, requiring limited supervision, and self-organized task decomposition is achieved through competition and specialization. The results are derived from computer simulations.

Lithium hydride powered PEM fuel cells for long-duration small mobile robotic missions

This paper reports on a study to develop power supplies for small mobile robots performing long duration missions. It investigates the use of fuel cells to achieve this objective, and in particular Proton Exchange Membrane (PEM) fuel cells. It is shown through a representative case study that, in theory, fuel cell based power supplies will provide much longer range than the best current rechargeable battery technology. It also briefly discusses an important limitation that prevents fuel cells from achieving their ideal performance, namely a practical method to store their fuel (hydrogen) in a form that is compatible with small mobile field robots. A very efficient fuel storage concept based on water activated lithium hydride (LiH) is proposed that releases hydrogen on demand. This concept is very attractive because water vapor from the air is passively extracted or waste water from the fuel cell is recycled and transferred to the lithium hydride where the hydrogen is “stripped” from water and is returned to the fuel cell to form more water. This results in higher hydrogen storage efficiencies than conventional storage methods. Experimental results are presented that demonstrate the effectiveness of the approach.

Tackling Learning Intractability Through Topological Organization and Regulation of Cortical Networks

A key challenge in evolving control systems for robots using neural networks is training tractability. Evolving monolithic fixed topology neural networks is shown to be intractable with limited supervision in high dimensional search spaces. Common strategies to overcome this limitation are to provide more supervision by encouraging particular solution strategies, manually decomposing the task and segmenting the search space and network. These strategies require a supervisor with domain knowledge and may not be feasible for difficult tasks where novel concepts are required. The alternate strategy is to use self-organized task decomposition to solve difficult tasks with limited supervision. The artificial neural tissue (ANT) approach presented here uses self-organized task decomposition to solve tasks. ANT inspired by neurobiology combines standard neural networks with a novel wireless signaling scheme modeling chemical diffusion of neurotransmitters. These chemicals are used to dynamically activate and inhibit wired network of neurons using a coarse-coding framework. Using only a global fitness function that does not encourage a predefined solution, modular networks of neurons are shown to self-organize and perform task decomposition. This approach solves the sign-following task found to be intractable with conventional fixed and variable topology networks. In this paper, key attributes of the ANT architecture that perform self-organized task decomposition are shown. The architecture is robust and scalable to number of neurons, synaptic connections, and initialization parameters.

Coevolving communication and cooperation for lattice formation tasks

Reactive multi-agent systems are shown to coevolve with explicit communication and cooperative behavior to solve lattice formation tasks. Comparable agents that lack the ability to communicate and cooperate are shown to be unsuccessful in solving the same tasks. The control system for these agents consists of identical cellular automata lookup tables handling communication, cooperation and motion subsystems.

Evolving multirobot excavation controllers and choice of platforms using an artificial neural tissue paradigm

Autonomous robotic excavation has often been limited to a single robotic platform using a specified excavation vehicle. This paper presents a novel method for developing scalable controllers for use in multirobot scenarios and that do not require human defined operations scripts nor extensive modeling of the kinematics and dynamics of the excavation vehicles. Furthermore, the control system does not require specifying an excavation vehicle type such as a bulldozer, front-loader or bucket-wheel and it can evolve to select for an appropriate choice of excavation vehicles to successfully complete a task. The “Artificial Neural Tissue” (ANT) architecture is used as a control system for autonomous multirobot excavation and clearing tasks. This control architecture combines a variable-topology neural-network structure with a coarse-coding strategy that permits specialized areas to develop in the tissue. Training is done in a low-fidelity grid world simulation environment and where a single global fitness function and a set of allowable basis behaviors need be specified. This approach is found to provide improved training performance over fixed-topology neural networks and can be easily ported onto different robot platforms. Aspects of the controller functionality have been tested using high fidelity dynamics simulation and in hardware. An evolutionary training process discovers novel decentralized methods of cooperation employing aggregation behaviors (via synchronized movements). These aggregation behaviors are found to improve controller scalability (with increasing robot density) and better handle robot interference (antagonism) that reduces the overall efficiency of the group.

An island model for high-dimensional genomes using phylogenetic speciation and species barcoding

A new speciation method for parallel evolutionary computation is presented, designed specifically to handle high-dimensional data. Taking inspiration from the natural sciences, the Phylogenetic Relations Island Speciation Model (PRISM) uses common ancestry and a novel species barcoding system to detect new species and move them to separate islands. Simulation experiments were performed on Multidimensional Knapsack Problems with different fitness landscapes requiring 100-dimensional genomes. PRISMs performance with various parameter settings and on the various landscapes is analyzed and preliminary results show that PRISM can consistently produce optimal or near-optimal solutions, outperforming the standard Genetic Algorithm and Island Model in all the performed experiments.

Application of Coarse-Coding Techniques for Evolvable Multirobot Controllers

Robots, in their most general embodiment, can be complex systems trying to negotiate and manipulate an unstructured environment. They ideally require an ‘intelligence’ that reflects our own. Artificial evolutionary algorithms are often used to generate a high-level controller for single and multi robot scenarios. But evolutionary algorithms, for all their advantages, can be very computationally intensive. It is therefore very desirable to minimize the number of generations required for a solution. In this chapter, we incorporate the Artificial Neural Tissue (ANT) approach for robot control from previous work with a novel Sensory Coarse Coding (SCC) model. This model is able to exploit regularity in the sensor data of the environment. Determining how the sensor suite of a robot should be configured and utilized is critical for the robot’s operation. Much as nature evolves body and brain simultaneously, we should expect improved performance resulting from artificially evolving the controller and sensor configuration in unison. Simulation results on an example task, resource gathering, show that the ANT+SCC system is capable of finding fitter solutions in fewer generations. We also report on hardware experiments for the same task that show complex behaviors emerging through self-organized task decomposition.

Hybrid Fuel Cells Power for Long Duration Robot Missions in Field Environments.

Mobile robots are often needed for long duration missions. These include search and rescue, sentry, repair, surveillance and entertainment. Current power supply technology limit walking and climbing robots from many such missions. Internal combustion engines have high noise and emit toxic exhaust while rechargeable batteries have low energy densities and high rates of self-discharge. In theory, fuel cells do not have such limitations. In particular Proton Exchange Membrane (PEMs) can provide very high energy densities, are clean and quiet. However, PEM fuel cells are found to be unreliable due to performance degradation. This can be mitigated by protecting the fuel cell in a fuel-cell battery hybrid configuration using filtering electronics that ensure the fuel cell is isolated from electrical noise and a battery to isolate it from power surges. Simulation results are presented for a HOAP 2 humanoid robot that suggests a fuel cell powered hybrid power supply superior to conventional batteries.

A Multiagent Methodology for Lunar Robotic Mission Risk Mitigation

The proposed return to the Moon by 2020 will represent one of the one of the most dramatic and challenging steps in human exploration as the international community prepares to establish a permanent presence, a homestead in the ultimate frontier. Prior to sending humans, however, there will be a number of robotic precursor missions. Even after humans alight on our closest celestial neighbor, robots will continue to play a crucial role, performing tasks that are too dangerous or even too mundane for astronauts. We must accordingly seek to mitigate mission risk whenever and wherever possible. Excavation tasks, for building landing pads, constructing habitats and generally establishing infrastructure, will undoubtedly be delegated to robotic systems. We propose that a multiagent methodology will be required to successfully accomplish these tasks and mitigate the associated risks. However, a multiagent approach in an unstructured environment will pose significant control challenges. We present a control architecture and philosophy for multiagent robotic systems. Such a system has been implemented in computer simulation and in a representative network of small laboratory rovers. The control paradigm is based on a flexible machine learning algorithm, which we call an “artificial neural tissue.” An evolutionary approach, that is, an artificial Darwinian selection process, is used to derive the control strategy in computer simulation. The result of this process can then be directly ported to the physical system to accomplish the desired tasks..