Automatic Engineering of General Robotic Problem Solvers

AutoFac

Abstract :  Robotics currently lacks fully autonomous capabilities, especially where task knowledge is incomplete and optimal robotic solutions cannot be pre-engineered. The intersection of evolutionary robotics, Artificial Life (ALIFE) and embodied artificial intelligence is a promising paradigm for generating general robotic problem-solvers suitable for adapting over extended periods (days to years) in unexplored, remote and hazardous environments. To address the automation of evolving robotic systems, we propose fully autonomous, embodied ALIFE factories, situated in any environment as general problem-solvers, where groups of robots (collective robotic systems) are designed and produced as solutions to given problems. Such ALIFE factories would be adaptive solution designers, producing fit-for-purpose groups of physical robots (designed in simulation with meta-heuristics), specifically designed and built to solve userassigned mission-objectives in new environments. Evolutionary machine-learning on robot controllers would enable discovery of necessary stepping-stones tasks towards accomplishing a pre-specified mission objective. Prerequisite technologies to realise such ALIFE factories, have already been experimentally demonstrated. Vast scientific and enterprise opportunities await in future applications such as asteroid mining, terraforming, space and deep sea exploration, though currently no suitable solution exists.

This research thus proposes a novel computational methodology for constructing ALIFE factories that enable the automated design and manufacture of problem-solving robots. Such ALIFE factories would thus be the designers and builders of autonomous collective robotic systems that could be deployed to solve any user-given problem in new environments. These ALIFE factories would be generalist problem-solvers, automatically designing robots (in simulation) and then physically producing the robots as specialist solutions in given environments.

Evolutionary Damage Control : Hexapod Case Study

Hexapod

Abstract :  Autonomous robots will potentially explore distant or hostile environments which would otherwise be entirely inaccessible, or pose a substantial risk to humans. This provides enormous benefits to scientific exploration, search and rescue, and disaster recovery. Due to the complexity and unpredictability of these harsh environments, hardware reliability has been identified as a significant challenge. This is particularly pertinent as in situ repair or retrieval is not possible. Humans and animals have evolved the ability to rapidly adapt to unexpected injuries and impediments. Robots with this ability could continue to function even when faced with unexpected hardware failures or damage.

This project uses a legged hexapod robot as a physical experimental evolutionary robotics platform for evaluating various online and offline automated controller adaptation methods. The objective is to adapt control behaviour to cope with morphological change (sensor and actuator damage and failure), such that the hexapod continues to function as effectively as possible. An end goal is to devise new methods that automate the adaptation of morphological robust controllers -- that is, highly plastic controllers that adapt behaviour to couple with morphological change (physical damage) for any autonomous robot, such that the robot can continue to reasonably operate in its environment, accomplishing tasks, despite damage sustained.  


  • Project Title : Evolutionary Damage Control : Hexapod Case Study
  • Funding : National Research Foundation : Human and Social Dynamics in Development (2019 - 2021) : Grant No: 118557.
  • Collaborators : Dr. L Raw, Department of Mechanical and Mechatronic Engineering, UCT;
    Prof. A Patel, Department of Electrical Engineering, UCT.

Social Complexity in Predynastic Egypt: An Agent-Based Approach

EgyptABM

Abstract : The origin and rise of complex states in antiquity has been a subject of considerable debate since the beginning of the modern discipline of archaeology. The case of ancient Egypt, one of the world's earliest and longest-lived examples of a pristine state, has long been a point of fascination for scholars, but without a clear consensus on how or why this state emerged when and where it did. Although there has been considerable advances in archaeological research in Predynastic Egypt in the past several decades, scholars still struggle to adequately narrate and understand this process. This is partially because of the fragmentary nature of the archaeological record, but also because of an inability to test and critically evaluate narrative models. What is needed are better analytical tools for developing, testing, challenging, and consequently improving our narrative models and reconstructions.

This project proposes to develop such a tool in the form of an Agent-Based Model (ABM), a type of computational simulation long favoured in the social sciences for its ability to study and analyse complex system behaviour, The model will be used to design experiments that examine the social dynamics of early Egypt, including the emergence of entrenched inequality, urbanism, social hierarchy, networks, and ideology of kingship. The goal is to demonstrate how the Egyptian state emerged as a result of the meaningful actions of individuals pursuing their own interests within the particular environmental conditions of the Nile Valley in the fourth millennium BC, as well as compare this system to similar case studies in social complexity in Africa more broadly. 


  • Project Title : The Emergence of Social Complexity and an Autocratic State in Predynastic Egypt: An Agent-Based Approach
  • Funding : National Research Foundation : Human and Social Dynamics in Development (2019 - 2021) : Grant No: 118557.
  • Collaborators : Dr. J Nitschke, Department of Ancient Studies, Stellenbosch University.