Towards Adaptive Robotic Green Plants

Fig. 1. Prototype of the cyborg, the final version will feature a casing hiding the electronics. Humans often use green plants as furniture in their built environments. More generally, the status of plants as living beings is highly dependent on their lack of mobility. The goal of this project, which is a collaboration between artists and roboticists, is to lead the general public to a questioning about the role of plants in the society through an artistic installation. We do so by endowing a green plant with motion, perception, learning and adaptation capabilities while retaining the emotional characteristic of a plant, such as calm and gently random behaviours. Several existing works involve green plants and robotics. For instance, [6] builds a " fake " plant mimicking the behaviour of a real one. In [2], a group of robots based on the iRobot Create manage a garden, caring for the plants. More into the artistic direction, [5] demonstrates a robotic flower pot moving imperceptibly slowly, and thus mimicking this particular attribute of green plants, that is, the slow reaction to the environmental changes. In this paper, the plant adapts its behaviour using a genetic algorithm. In this direction, previous works have emphasized the importance of adaptation [4] and the implied " play " behaviour [1] to create interesting robots from an artistic point of view. At the level of interaction, works have shown that an artistic robot should have its own life, avoiding or ignoring humans [3]. In this work, we propose a cyborg, that is, an agent combining biological and technological elements. This cyborg consists of an iRobot Create, a computer running Linux, a normal plant and additional sensors (Fig. 1). The final version will also feature a casing hiding the electronics. The cyborg lives its own life, following its internal needs of water, sunlight and electrical energy. It satisfies these needs by performing actions to fetch water, to find sunny locations and to recharge the robot's battery. To do so, and in contrast with existing works, the robot employs a well-founded probabilistic planning algorithm. This algorithm takes as input the current state of the cyborg needs, and produces as output a sequence of actions.