unit-code
Contemporary knowledge workers in the 21st century are becoming increasingly global and nomadic leading to feelings of social separation and loneliness. In reaction to this, BEAST is designed as an autonomous spatial entity which interacts with the changing social and natural conditions of its environment to catalyse new playful social spaces. The body plans of BEAST extend the characteristics of scissor structures and are composed of linkages of various tensile and compression components with passive and active joints enabling them to maximise degrees of adaptation while minimising actuation. BEAST is a cyber-physical system made self-aware through a lidar sensory system and through motor sensor feedback, evolving its self-adapting behaviour in relation to its changing human and non-human environment. The research involves parallel development and integration of robotic prototypes with sensor actuation control systems, and a Unity-based simulator using reinforcement learning to learn adaptive policies for mobility, spatial configuration, and shading. Throughout the course each day BEAST senses its internal structural state in relation to the location and movements of people and the sun, continuously adjusting its configuration in a dynamic balance to provide evolving social spaces.
The Beast moves in real time by sensing sunlight and people.
Used to customise the Beast interface. Users can pick and choose the components they need to generate a viable Beast.
Using radar scans, the Beast can store its environment as a point cloud.
The connection between the prototypes is represented in detail with geometric diagrams to complete the 2D to 3D conversion.
The bounding box is used instead of the rod, and a reasonable connection form is selected. Rods are connected from Component to Component Cluster and finally to BEAST.
The library of BEAST's rods and bounding boxes.
Different functions of different scales of BEAST combined transformation in different environments.
Develop joint systems of different generations through experiments to cope with diversified morphological changes.
The entire structure is deformed by rotating a single joint.
Beast body plans are composed of linkages of various components with passive and active joints enabling them to maximise degrees of adaptation while minimising actuation.
Beast is self-aware through a lidar sensory system and through motor sensor feedback, evolving its self-adapting behaviour in relation to its changing human and non-human environment.
The BEAST 02 is composed of 3 kinds of components and 66 rods, and it is equipped with LiDAR sensor system to become a self-aware BEAST.
Simulations of the five basic elements, as well as tests of their variations.
The geometry analysed in the previous stage was set up in a virtual environment, and gravity, material collision and external force were added.
BEAST 03 training parameters and a comparison of pre-training and post-training behaviour.
From Component evolution to BEAST reinforcement training reports, environment settings, training rewards, and agent learning are all shown in the reports.
The movement record of the agent. As the composition of components increases, the movement ability of the agent becomes more complex.