Anirudh Tiwathia

Bio [2004]
Anirudh Tiwathia is starting his senior year at Vassar College. He is a double major in cognitive science and computer science, and is interested in a wide variety of scientific and philosophical questions, such as the Principle of Computational Equivalence. His research so far has been in cognitive science.

Project Title
Using 2D Turing Machines to Create Abstract Models of Behavior-Based Robots

Project
The initial approaches to robotics were based around providing robots with the ability to create world-models based on both sensory input and previously encoded knowledge about the world. These robots then chose the appropriate actions by combining predetermined plans with this world-model. Such robots proved extremely ineffective at performing even the most basic of tasks. Thus this approach was eventually discarded. The two main contributions to the movement away from the world-model-based paradigms toward behavior-based robotics were led by Rodney Brooks's work at MIT and Valentino Braitenberg's book "Vehicles." Braitenberg demonstrated how simply wired robots could give rise to some complex-appearing behaviors, whereas Brooks emphasized the use of subsumption architectures. Instead of concentrating on building-in complexity of behavior, behavior-based robotics is a paradigm based on providing the robot with simple hierarchies of appropriately chosen simple localized "rules" or "behaviors." The interaction of these simple, low-level, local rules and the resulting interactions with the environment is what gives rise to the complexity in observed behavior of the robots. Robots designed on behavior-based approaches tend to act in ways similar to behavior found in biological systems. In fact, these robots have been used to model different kinds of insect behavior such as, for example, ant-colony foraging behavior. Such an approach is gaining greater popularity for the purposes of modeling animal behaviors. Moreover, it is having a significant impact on theories of cognitive science relating to perception and action loops in various biological systems.

The emphasis on providing only very simple, low-level rules quite obviously ties behavior-based approaches to the "NKS way of thinking," since both formalisms believe in achieving complex behavior from low-level, local specifications. Moreover, within subsumption architectures, usually the immediate behavior is determined by only one active state. The robot cycles through the various internal states as its behavior interacts with the environment. The idea of cycling through a set of discrete states makes 2D Turing machines (TMs) an ideal choice in modeling behavior-based robots.

Within this summer school, I would like to model a 2D Turing machine to create an abstract model of a simple behavior-based robot. By providing the Turing machine with appropriately chosen local rules, the TM will be able to move around in its environment and perform goal-directed behavior. Initially, it would be helpful to familiarize myself with various 2D Turing machines and their behavior. MAA Online and MathWorld are both rich sources for a variety of 2D Turing machines. Starting from there, the goal would be to develop a behavior-based model. For example, one particular behavior-based model could be a Turing machine that performs random walks across its 2D space and then collects all the black dots encountered into one side of the grid.

Favorite two-color, radius-2 rules
Rules chosen: 837301587, 3528706269, 3494508071, 167508322