Evolved Control of Natural Plants

Abstract

Mixing societies of natural and artificial systems can provide interesting and potentially fruitful research targets. Here we mix robotic setups and natural plants in order to steer the motion behavior of plants while growing. The robotic setup uses a camera to observe the plant and uses a pair of light sources to trigger phototropic response, steering the plant to user-defined targets. An evolutionary robotic approach is used to design a controller for the setup. Initially, preliminary experiments are performed with a simple predetermined controller and a growing bean plant. The plant behavior in response to the simple controller is captured by image processing, and a model of the plant tip dynamics is developed. The model is used in simulation to evolve a robot controller that steers the plant tip such that it follows a number of randomly generated target points. Finally, we test the simulation-evolved controller in the real setup controlling a natural bean plant. The results demonstrate a successful crossing of the reality gap in the setup. The success of the approach allows for future extensions to more complex tasks including control of the shape of plants and pattern formation in multiple plant setups.

Social Integrating Robots Suggest Mitigation Strategies for Ecosystem Decay

“…Two general approaches were followed, different in scale (in space and time) and precision. (1) A system consisting of a single board computer with a camera and control over two light sources together with a single freshly sprouted bean plant was used to guide the growing shoots to multiple targets in space using image detection and machine learning (detailed in Hofstadler et al, 2017). In these experiments, it typically took the bean shoot 2-3 days to grow out of the space monitored by the camera, corresponding to ∼50 cm of bean shoot.…”

“…We showcase the model mimicking the behaviour of the closedloop bean tip controllers artificially evolved in Hofstadler et al (2017;Figure 11). The task is to guide a single growing and nutating tip through specific targets on the 2D plane of the camera projection during its (growth-)journey through the image.…”

Constructing living buildings: a review of relevant technologies for a novel application of biohybrid robotics

“…Ad hoc approaches to this problem (e.g. [109,110]) construct a data-driven model by image processing time-lapse records of a certain species in a given set-up, from a few initial experiments. More generalized approaches could be investigated, building from a variety of models in plant science literature.…”

“…Research on steering the morphological development of individual plants is rare, as agricultural concerns, for instance, do not motivate such studies. However, there is a line of research on shaping plants that develops an automated process of evolving controllers that direct the growth of a single plant to certain goals [109,110,162,163]. Machine vision was used to understand the behaviour of single bean plants in reaction to external light stimuli, and to construct data-driven models of the plant’s growth and motion.…”

“…The behaviours of plants that can be interfaced for robotically steered control are reviewed in §2. Centralized robotic control of plant stimuli is explored by Wahby et al [109,162], Hofstadler et al [110] and Wahby et al [163], using a purpose-specific model of plant growth combined with controllers evolved in simulation to predictably steer growth to two-dimensional geometric targets. In this set-up, the plant has no mechanical scaffold, but the height to which it can support itself is not tall enough for building-sized growth.…”