Robotic Sensing and Stimuli Provision for Guided Plant Growth

1. Plant species selection procedure

NOTE: This protocol focuses on the plant behaviors related to climbing, directional responses to light, and the health and survival of the plants in the specific season, location, and experimental conditions.

1. Select a plant species known to display strong positive phototropism^15,16 towards UV-A and blue light (340–500 nm) in the growing tips.
2. Select a species that is a winder, in which the circumnutation¹⁷ behavior is pronounced and the growing tip has helical trajectories with large enough amplitude to wind around the mechanical supports used in the specific experimental conditions. The twining¹⁸ behavior exhibited by the selected winder should tolerate the environment and nutrient conditions present in the experiment and should tolerate mechanical supports with angle of inclination up to 45°.
3. Select a species that will grow reliably and quickly in the experimental conditions, with an average growth speed not less than approximately 5 cm per day, and preferably faster if possible.
4. Select a species that will display the required behaviors in the present season and geographic location.
5. Ensure the species tolerates the range of environmental parameters that will be present in the experimental setup. The plant should tolerate an absence of green light and an absence of light outside the visible spectrum (400–700 nm). The plant should also tolerate any present fluctuations in temperature, kept at approximately 27 °C, as well as any present fluctuations in humidity and watering.

2. Robot conditions and design

1. Organize robot capabilities into decentralized nodes with single-board computers (see Figure 1 and Figure 2), integrated into modular mechanical supports. Ensure each identical robot node is able to control and execute its own behavior.
2. For robotic provision of stimuli to plants, provide blue light (400–500 nm) to plants at controllable intervals, at an intensity that will trigger their phototropic response, from the direction and orientation required for the respective portion of the experiment.
   1. Select a red-green-blue (RGB) light-emitting diode (LED) or an isolated blue LED. In either case, include an LED with a blue diode with peak emission ƛ_max = 465 nm.
   2. Select an LED that when congregated in groups and set in the precise conditions of the utilized robot can maintain the required light intensity level in each direction tested in the experiment setup. For each direction being tested, ensure that the blue diodes in the LEDs in a single robot are collectively capable of maintaining a light intensity level of approximately 30 lumens without overheating, when situated in the utilized robot enclosure and any utilized heat dissipation strategies. The selected LED should have a viewing angle of approximately 120°.
      NOTE: For example, in a robot utilizing three LEDs per direction, with microcontroller-enabled regulation of intensity, if the blue diodes emit with maximum light intensity Φ = 15 lumens, then without overheating they should be able to maintain 65% of the maximum.)
   3. Interface the LEDs to the robot’s single-board computer, via LED drivers that regulate the power supply according to the required brightness. Enable individual control, either of each LED or of the LED groups serving each direction being tested in the setup.
3. For the sensing procedure for the proximity of plant growing tips (see Figure 3B), use processed readings from infrared proximity (IR-proximity) sensors to reliably and autonomously detect the presence of plants approaching from each direction tested in the setup.
   1. Select an IR-proximity sensor that regularly detects the growing tip of the selected plant species, when arranged perpendicularly to the central axis of the direction from which the plant approaches, as tested in an unobstructed environment. Ensure successful detection occurs starting from a distance of 5 cm, as seen in Figure 3A starting at the timestamp labelled ‘07.04.16’ on the horizontal axis.
   2. Interface each IR-proximity sensor to the robot’s single-board computer, and implement a weighted arithmetic mean approach to process the sensor readings into a determination of whether a plant is present within 5 cm. Use the sensor readings from the most recent five s to give 20% of the final average weight used in detection.
   3. Ensure the selected IR-proximity sensor does not emit critical wavelengths that could interfere with the light-driven behaviors of the selected species. Ensure that wavelengths emitted by the sensor below 800 nm are not present at distances greater than 5 mm from the sensor’s IR source, as measured by spectrometer.
4. Distribute the experiment functions over the set of robots, such that each robot can autonomously manage the portions that proceed in its own local area. Arrange the robots’ provision of light stimuli and sensing capabilities according to the respective plant growth directions being tested.
   1. Compose each robot around a single-board computer that is Wireless Local Area Network (WLAN) enabled. Interface the computer to sensors and actuators via a custom printed circuit board (PCB). Power each robot individually, with its own battery backup.
   2. Include one IR-proximity sensor per direction being tested for approaching plants, according to the above requirements.
   3. Include enough LEDs to deliver the above blue light requirements, per direction being tested for approaching plants.
      1. If using RGB LEDs rather than blue LEDs, optionally enable emittance from the red diode when the blue diode is not in use, to augment the red light delivery described below (for plant health via the support of photosynthesis).
      2. If red light is emitted from the robots at certain intervals, then use red diodes with peak emission at approximately ƛ_max = 625–650 nm, with no critical wavelengths overlapping the green band (i.e., below 550 nm) or the far-red band (i.e., above 700 nm).
      3. Do not allow red diodes to produce heat levels higher than those of the blue diodes.
   4. Include hardware that enables local cues between robots. Include a photoresistor (i.e., light-dependent resistor or LDR) for each direction of a neighboring robot to monitor their light emittance status. Alternatively, communicate the status of local neighbors via WLAN.
   5. Include hardware to dissipate heat, as required by the conditions of the selected blue diodes and the utilized robot enclosure. Execute by a combination of aluminum heatsinks, vents in the robot’s case enclosure, and fans. Activate fans by a digital temperature sensor on the single-board computer or supplemental PCB.
   6. Organize the robot components such that the relevant directions are uniformly serviced.
      1. Position the blue diodes to distribute an equivalent light intensity to each of the directions from which plants may approach (i.e., from the mechanical supports attached to the robot’s lower half, see step 2.5). Orient each diode in the robot case such that the center axis of its lens angle is within 60° of each axis of mechanical support it services, and position it to not be blocked by the robot case.
      2. Position the IR-proximity sensors equivalently for their respective approaching growth directions (i.e., from the mechanical supports attached to the robot’s lower half, see step 2.5). Position each IR-proximity sensor within 1 cm of the attachment point between the robot and the mechanical support being serviced, and orient it such that its viewing angle is parallel to the support axis. Ensure its emitter and receiver are not blocked by the robot case.
      3. Position any photoresistors for local communication equivalently for each direction facing a neighboring robot in the setup (i.e., from all mechanical supports attached to the robot, see 2.5). Orient each photoresistor such that the center axis of its viewing angle is within 45° of the support axis it services, and position to not be blocked by the robot case.
   7. Assemble all components with the single-board computer (refer to the block diagram in Figure 2). Ensure the computer can be easily accessed for maintenance after assembly.
      1. Interface LEDs to the computer via LED drivers using pulse width modulation. Use a fixed mechanical connection between the LEDs and either the case or the heatsink, and use a mechanically unconstrained connection between the LEDs and the computer.
      2. Interface fans to the computer via linear regulator (i.e., switch) using a general purpose input/output header pin. Affix fans where adequate airflow is available, while also ensuring no mechanical stress is placed on them.
      3. Interface IR-proximity sensors and photoresistors via analog-to-digital converter, using serial peripheral interface. Use a fixed mechanical connection from the sensors to the case, and a mechanically unconstrained connection to the computer.
      4. Manufacture the robot case from heat-resistant plastic using either selective laser sintering, stereolithography, fused deposition modelling, or injection molding.
5. Integrate the robots into a set of modular mechanical supports that dually hold the robots in position and serve as climbing scaffold for the plants, restricting the plants’ likely average growth trajectories. Design the robots to serve as supplemental mechanical joints between the supports, positioned such that they intersect the plant growth trajectories.
   1. Minimize the size of the robot, and ensure that it can be reliably surpassed by an unsupported growing tip of the selected plant species. Reduce robot size to the greatest extent possible to increase experiment speed.
   2. Shape the external walls of the robot body to be as unobtrusive to plant growth as possible when a growing tip incrementally navigates around the robot. Round or facet the robot body to not block the helical trajectory of circumnutation in twining plant species. Exclude sharp protrusions and acute indentations.
   3. Select a material and profile (i.e., shape of cross-section) for the mechanical supports, such that the selected plant species can effectively climb it, for instance a wooden rod with circular profile of a diameter roughly 8 mm or less. Ensure the mechanical supports are structurally stiff enough to support the plants and robots within the setup, augmented by a transparent acrylic sheet behind the setup.
   4. On each robot include attachment points to anchor the specified mechanical supports. Include one for each direction by which a plant may approach or depart a robot.
      1. For each attachment point, include a socket in the robot case, with dimensions matching the cross-section of the support material.
      2. Set the socket with a depth not less than 1 cm. Keep the socket shallow enough that the support does not collide with components inside the robot.
   5. Arrange the mechanical supports in a regularly gridded pattern, uniformly diagonal with an angle of inclination at 45° or steeper. Make the lengths of the supports uniform. The minimum exposed length of the support is 30 cm, to allow sufficient room for the climbing plants to attach after exploring the area in their unsupported condition. The preferred exposed length is 40 cm or more, to allow some buffer for statistically extreme cases of plant attachment.
   6. Assemble the mechanical elements with the robots. The following protocol assumes an exposed support length of 40 cm, and a setup of eight robots in four rows (see Figure 6). For other sizes, scale accordingly.
      1. On the floor surface, build a stand 125 cm wide that is capable of holding the setup in an upright position.
      2. Affix a 125 cm x 180 cm sheet (8 mm thick or more) of transparent acrylic to the stand, such that it stands upright.
      3. Position pots with the appropriate soil on the stand, against the acrylic sheet.
      4. Affix two mechanical y-joints to the acrylic sheet, 10 cm above the pots. Position the joints 45 cm and 165 cm to the right, respectively, of the left edge of the stand.
      5. Affix two supports to the left y-joint, leaning 45° to the left and to the right, and affix one support to the right y-joint, leaning 45° to the left.
      6. Affix two robots to the acrylic sheet, and insert the ends of the previously placed supports into the sockets in the robot cases. Position the robots 35 cm above the y-joints, and 10 cm and 80 cm to the right, respectively, of the left edge of the stand.
      7. Repeat the pattern to affix the remaining robots and support in the diagonally gridded pattern (see Figure 6), such that each row of robots is 35 cm above the previous row, and each robot is horizontally positioned directly above the robot or y-joint that is two rows beneath it.

3. Robot software

1. Install an operating system (e.g., Raspbian) on the single-board computers of the robots.
2. During each experiment, run the software protocol on each robot in parallel, enabling their distributed autonomous behavior (see Wahby et al.¹⁴, for pseudocode and more details).
3. Establish two possible states for the robot: one being a stimulus state during which the robot emits blue light at the intensity described above; the other being a dormant state during which the robot either emits no light or emits red light as described above.
   1. In stimulus state, send a pulse width modulation (PWM) signal via the single-board computer with a frequency corresponding to the required brightness to the blue LEDs drivers.
   2. In dormant state, trigger no LEDs, or if needed send a PWM signal to only the red LEDs drivers.
4. In control experiments, assign all the robots the dormant state.
5. In single-decision experiments, assign one robot the dormant state and one robot the stimulus state.
6. In multiple-decision experiments, start the Initialization process, as follows.
   1. Supply to each robot a full configuration map of the pattern of plant growth to be tested in the current experiment.
   2. Set the location of the robot within the pattern, either automatically using localization sensors or manually.
   3. Compare the robot’s location to the supplied map. If the robot’s location is the first location on the map, set the robot to stimulus; otherwise, set the robot to dormant. Initialization process ends.
7. In multiple-decision experiments, start the Steering process, as follows. Execute iteratively.
   1. Check the robot’s IR-proximity sensor reading to see if a plant has been detected.
   2. If a plant is detected and the robot is set to dormant, then maintain.
   3. If a plant is detected and the robot is set to stimulus, then:
      1. Notify the adjacent neighboring robots that a plant has been detected, and include the robot’s location in the message.
      2. Set the robot to dormant.
      3. Compare the robot’s location to the map. If the robot is at the last location on the map, then send a signal over WLAN that the experiment is complete.
   4. Check the robot’s incoming messages from its adjacent neighboring robots to see if one of them that was set to stimulus has detected a plant.
   5. If a stimulus neighbor has detected a plant, compare that neighbor's location to the robot’s location, and also compare to the map.
   6. If the robot is at the subsequent location on the map, set the robot to stimulus.
   7. End the iterative loop of the Steering process once a signal has been received that the experiment is complete.

4. Plant health monitoring and maintenance procedure

1. Locate the experiment setup in controlled environmental conditions—specifically, indoor with no incident daylight or other light external to the conditions described below, with controlled air temperature and humidity, and with controlled soil watering. Monitor the conditions with sensors that are connected to a microcontroller or single-board computer that is WLAN enabled.
2. Maintain plant photosynthesis using LED growth lamps external to the robots and facing the experiment setup.
   1. Use the growth lamps to deliver monochromatic red light to the setup, with red diodes having peak emission at approximately ƛ_max = 625–650 nm, with no critical wavelengths outside the range 550–700 nm, except for a low incidence of ambient blue light if helpful for the health of the selected species. If a low incidence of ambient blue light is included, restrict to levels at a very minor fraction of those emitted by a single robot.
   2. Provide the levels of red light required for the health of the selected species, usually roughly 2000 lumens or more in total.
   3. Orient the growth lamps to face the experiment setup, such that their emittance is distributed roughly evenly over the growth area.
   4. Monitor the ambient light conditions using an RGB color sensor.
3. After germinating, provide each plant its own pot at the base of the experiment setup. Provide suitable soil volume and type for the selected species. Ensure the soil and seeds have been sanitized prior to germination. Use suitable pest control methods to prevent or manage insects if present.
4. Regulate air temperature and humidity levels, accordingly for the selected species, using heaters, air conditioners, humidifiers, and dehumidifiers. Monitor levels using a temperature-pressure-humidity sensor.
5. Monitor the soil using a soil moisture sensor. Maintain an appropriate rate of watering for the selected species. Execute using an automated watering system where water is delivered to the soil via nozzles as triggered by the soil moisture sensor readings, or water soil manually, as regulated by the sensor readings.

5. Experiment design

1. Place robots and mechanical supports in a grid large enough to cover the growth area and pattern being tested in the experiment, not smaller than one row and two columns of robots.
2. Below the bottom row of robots, place a row of the standard diagonal mechanical supports, matching those throughout the setup. Where the lower ends of these supports intersect, join them mechanically with a ‘y-joint.’ For each ‘y-joint’ at the base of the setup, plant a uniform number of plants according to the size of the diagonal grid cell (roughly one plant per 10 cm of exposed mechanical support length), with the plant health maintenance conditions described above.
3. Select an experiment type to run, and where relevant select a quantity and distribution of robots.
   1. Experiment type 1: Control
      NOTE: This experiment type tests growth of the climbing plants in conditions absent of light stimuli to trigger phototropism. It can run on any size and shape of setup.
      1. Assign all robots the dormant state (see step 3.4) and run continuously until results are manually assessed to be complete.
      2. Observe whether plants attach to the mechanical supports. In a successful experiment, none of the plants will find or attach to the mechanical supports.
   2. Experiment type 2: Single-decision
      NOTE: This experiment type tests the plants’ growth trajectories when presented with binary options—one support leading to a dormant robot and one support leading to a stimulus robot. It runs only on the minimum setup (i.e., one row, two columns).
      1. Assign one robot the dormant state (see 3.5) and one robot the stimulus state. Run continuously until one of the two robots detects a plant with the IR-proximity sensor.
      2. Observe plant attachment to mechanical support, growth along support, and sensor readings of stimulus robot. In a successful experiment, the robot with the stimulus state will detect a plant after it had grown along the respective support.
   3. Experiment type 3: Multiple-decision
      NOTE: This experiment type tests the plants’ growth when presented with multiple subsequent stimuli conditions, that trigger a series of decisions according to a predefined global map. It can run on any size and shape of setup that has more than the minimum number of rows (i.e., two or more).
      1. Provide to the robots a global map of the pattern to be grown (see steps 3.6-3.7.7).
      2. Observe the plant attachment events and pattern of growth along the mechanical supports.
         1. In a successful experiment, at least one plant will have grown on each support present in the global map.
         2. Additionally, in a successful experiment no plant will have chosen the incorrect direction when its growing tip is located at the currently active decision point.
         3. Do not consider extraneous growing tips here, if for instance a branching event places a new growing tip at an obsolete location on the map.

6. Recording Procedure

1. Store data from sensors and cameras initially at the single-board computer where the data has been generated onboard. Run onboard reply servers that respond to needed requests, such as the last stored sensor reading. At regular intervals upload the data and log files over WLAN to a local network-attached storage (NAS) device.
2. Capture time-lapse videos of the experiments continuously using cameras positioned at two or more vantage points, with at least one camera view encompassing the full experiment setup. Ensure the captured images are of high enough resolution to adequately capture the movements of the plant growing tips, typically only a few millimeters in width.
   1. Automate the image capture process to ensure consistent time intervals between captures, using an onboard camera on a single-board computer or a stand-alone digital camera that is automated with an intervalometer. Install lamps to act as flashes, automated similarly to the cameras. Ensure the flashes are bright enough to compete with the red light of the growth lamps without dramatically post-processing the images for color correction.
   2. Locate the flashes such that the experiment setup can be fully illuminated and therefore clearly visible in images. Synchronize the cameras and the flashes such that all cameras capture images simultaneously, during a 2 s flash period. Capture the images every 2 minutes, for the duration of each experiment.
3. Log the environmental sensor data, specifically the readings from the temperature-pressure-humidity sensor, the RGB color sensor, and the soil moisture sensor. Log the data from all robots in the setup, specifically the IR-proximity sensor and photoresistor readings, as well as the internal state of the robot which defines its LED emittance status.
4. Make all recorded data available for remote monitoring of the experiments, via regular real-time reports, to ensure correct conditions are maintained for the full experiment duration up to several months.