The hardware design of the robots usually reflects its purpose. For example, leg for walking, arm for reaching, hand for grasping, etc. The function of mechanical systems is explicitly defined by designers and the simple shape consisting of straight lines or circular arcs is reasonable for its clarified function. On the other hand, our real environment is full of diverse shapes. The complex shape has been formed by various factors, biological growth, aging, weathering and so on. Utilizing diverse objects to create robots, we might be able to see behaviors and motions we’ve never seen. We aim to find unpredictable motions we could not discover with simple shapes. In this work, as a primitive function of the robot, we focus on stand-up behavior. For the materials that bricolage the robot, we select tree branches as objects with diverse shapes. The robots' poses are generated aiming to maximize the height of the bodies. Robots with diverse body shapes change their pose by repeating trial and error in real time. Through the process of learning, this work portrays new functions and meanings given to commonplace objects. size: 3×3×1.5 m year: 2019 contents: robots, control unit, sensing camera

The hardware design of the robots usually reflects its purpose. For example, leg for walking, arm for reaching, hand for grasping, etc. The function of mechanical systems is explicitly defined by designers and the simple shape consisting of straight lines or circular arcs is reasonable for its clarified function. On the other hand, our real environment is full of diverse shapes. The complex shape has been formed by various factors, biological growth, aging, weathering and so on. Utilizing diverse objects to create robots, we might be able to see behaviors and motions we’ve never seen. We aim to find unpredictable motions we could not discover with simple shapes. In this work, as a primitive function of the robot, we focus on stand-up behavior. For the materials that bricolage the robot, we select tree branches as objects with diverse shapes. The robots' poses are generated aiming to maximize the height of the bodies. Robots with diverse body shapes change their pose by repeating trial and error in real time. Through the process of learning, this work portrays new functions and meanings given to commonplace objects.

Software The robot updates its motion according to the target behavior. A depth sensor installed above it is measuring the height of the robot's Center of Volume (CoV) from the ground. The purpose of the robot is to maximize the height of this CoV. We defined the optimized motion based on this criterion as standing. The robot changes its pose and updates input values to each motor through trial and error in real time. CMA‒ES, which is one of the evolutionary algorithms, is used for the generation of robot's poses and optimization. Through the process of optimization, we discover new functions and meanings given to branches of trees with various shapes. Hardware Various shape branches are collected to serve as the robot body and connected to servo motors to create the robot. The branches and servo motors are connected by connectors printed by a 3D printer. We designed several robots with different DoF in each.

Robot: Branch, 3D printed part. Servo motor 0.5 × 0.5 × 0.5 m 1.0-1.5 kg Control unit: Control PC, microcomputer 0.5 × 0.5 × 0.2 m 2.0 kg Sensor: Depth camera 0.1 kg Overall work size: 3 x 3 x 1.5 m (depth camera installed on frame or ceiling or tripod)