This simulation is intended to demonstrate communication between ACT-R-controlled agents that are embedded in a virtual environment. In the first phase, a teacher robot explores the maze and learns about its structure. In a subsequent phase, the teacher robot uses its knowledge of the maze to guide the movements of a student robot. The teacher and student are both implemented as ACT-R models that communicate with one another via the virtual environment – the student relays information about its perceptual environment to the teacher, and in response, the teacher issues movement-related instructions to the student.

The post ACT-R Robots – ‘H’ Maze Simulation 4 appeared first on Cognosys Ltd..

In this third simulation using the ‘H’ Maze task, we attempt to optimize the way that ACT-R models create cognitive maps for a virtual spatial environment. In particular, the creation of nodes (or place fields) in the map is limited to situations where the robot is faced with a decision that affects its directional heading. This constraint results in a much more efficient use of the robot’s mnemonic resources: the robot only attempts to remember those spatial locations that could actually have an impact on future navigation-related decisions.

The post ACT-R Robots – ‘H’ Maze Simulation 3 appeared first on Cognosys Ltd..

In this second simulation using the ‘H’ Maze task, an ACT-R-controlled virtual robot is given the ability to discriminate between visually-similar locations in the maze based on its proximity to visual landmarks. This enables the robot to avoid the kind of errors in navigational decision-making that were encountered in the first simulation.

The post ACT-R Robots – ‘H’ Maze Simulation 2 appeared first on Cognosys Ltd..

In this video, a virtual robot attempts to learn about the structure of a simple maze. The robot initially explores the maze and memorizes specific locations based on a combination of visual and tactile input. The robot has four eyes oriented in the forward, backward. left and right directions. It also has four whiskers that detect contact with the walls of the maze and a compass that provides directional information. An ACT-R cognitive model implements the spatial learning and spatial reasoning capabilities of the robot. The same model also controls the robot’s movements in response to sensory input.

The post ACT-R Robots – ‘H’ Maze Simulation 1 appeared first on Cognosys Ltd..

In this video, a virtual vehicle is tasked with the collection and removal of target objects (i.e., blue spheres) from a partially enclosed arena. The vehicle uses phototactic responses to locate the targets, and it then moves to a goal location. The goal location is represented by an opening in one of the walls of the arena and is illuminated by a red light. Once at the goal location, the vehicle disposes of the target by pushing it through the opening. In addition to visual information, the vehicle’s behaviour is guided by both tactile and proprioceptive information. All aspects of the vehicle’s behaviour are controlled by an ACT-R model. The model receives sensory information from the virtual environment, reasons over this information, and then makes decisions regarding behavioural output.

The post Blue Marble Simulation appeared first on Cognosys Ltd..

Video trailer for the Vehicles project. Shows various kinds of vehicle behaviour, including simple phototactic responses, obstacle avoidance, social interaction and conditioning.

The post Vehicles – Project Trailer appeared first on Cognosys Ltd..