11% savings on maintenance staff after building an automatic plant watering robot

API

AWS

IoT

Innowise’s robotics department developed a proprietary smart robot to navigate office premises and water plants without human intervention.

Customer

Industry

IT, Custom software development

Region

Client

Innowise

Innowise is a global full-cycle software development provider with 1500+ IT specialists on board. Our company renders turnkey software development services, delivering 850+ projects for customers from 30 countries worldwide.

Challenge: Innowise's in-house project to showcase our robotics expertise

Robotics is one of the ever-expanding trends in modern IT realities. Digital networks and artificial intelligence grow exponentially, given the rapid technological advancements in these fields.

With a focus on utilizing emerging technologies, Innowise adopts advanced solutions as they come to market. As proof of our domain excellence, our robotics department crafted a full-fledged autonomous robot from scratch to assist employees with watering plants. In this proprietary project, we showcased our robotic expertise to customers looking for IoT-driven solutions to automate routine tasks and eliminate human oversight.

Solution: Self-navigating robot watering plants with advanced elevating system

Our vetted robotics developers built an IRIS (Innowise Robotics Irrigation System) - an autonomous self-navigating robot for IoT plant watering in office premises. Besides implementing software such as SLAM, ROS, and LiDAR, we also built hardware, including a moving platform, water storage tank, and elevating system.

Mapping

Our robotics experts started by mapping office spaces to create a detailed IoT plant monitoring system, identifying the plants’ locations, obstacles, furniture, and other objects that may affect the robot’s movement. We ensured predictable and hassle-free routing across office rooms by utilizing SLAM technology, which simultaneously determines the robot’s location and creates an environment map using computer vision algorithms, LiDAR (laser scanners), and other sensor tools.

Our robotics specialists used LiDAR connected to the Raspberry PI microcomputer mounted directly on the robot to detect obstacles and identify plants. ROS (Robotic Operating System) and the main computer use this visual information to process navigation data, make route calculations, and map the office surroundings.

During this stage, our team faced the challenge of limited visibility in detecting plain objects like tables, shelves, chairs, and other interior items that restrict the robot’s view or can be misidentified. Additionally, we had to deal with dynamic obstacles in an office environment since employees and moving objects suddenly change positions and directions, forcing the robot to make instant decisions to avoid collisions. Our project team used computer vision and machine learning algorithms to address this issue, including image segmentation, object detection, noise filtering, and other methods. Also, we equipped our autonomous assistant with motion planning algorithms such as Rapidly-exploring Random Trees (RRT) and A* (A-star), which considers the position and shape of obstacles in identifying the optimal path in real-time.

Plant detection and QR codes

The project’s primary goal was training the robot to identify and locate objects on a map. Initially, we planned to use stereoscopic cameras to determine the plants’ location, calculate their position, and create a route. As a result of the brainstorming sessions, we devised an alternative scheme where the robot took a picture and recorded its coordinates in space. Robotics engineers used a neural network to find the plant in the frame, calculate its bounding box, and determine the flower’s direction.

As part of image processing projects, bounding boxes serve as reference points for object detection and create collision boxes for them. Based on the robot’s coordinates, the camera’s orientation, and the flower’s location, we drew a ray connecting the robot’s position with the plant. Upon repeating this process many times, we obtained many rays intersecting at one point and detecting the plant that needed watering.

Our engineers relied on models trained on COCO and ImageNet datasets to identify flowers in pots seamlessly. Based on this model, we filtered out all unnecessary classes and developed a custom detector that synchronizes the bounding box direction with the robot coordinates. To determine the precise spatial coordinates of the watering rod, we used a bundle of cameras and LiDAR.

Once the robot detects the plant, it should identify its accurate position in space and determine whether it should be watered. For this purpose, we labeled all the office pots with QR codes connected to databases where the watering history of all plants is kept.

Moving platform

Regarding hardware, the robotics team opted for a modular system, which included a moving platform containing electronics, a water storage tank, a battery, and a two-level elevator system. We used the aluminum profile of the V-Slot format to assemble the robot’s frame for its durability and lightness, enabling improved maneuverability and reduced energy consumption.

Instead of standard differential drives, we implemented omni-wheels at the corners of the robot to ensure smooth navigation. Omni-wheels, or omnidirectional wheels, present small disks (rollers) around the circumference that can rotate on their axis or perpendicularly, driving the entire system easily. In this way, the robot can move in any direction without rotating the main structure, using only the difference in velocity between each wheel.

Elevator and irrigation

Flowers are displayed on employees’ desks, shelves, racks, high bookcases, and other places that are hard to access for employees. Rather than building a high robot, our experts assembled a lifting mechanism based on sliding rollers, eliminating the need for labor-intensive and economically inefficient bookcase-high construction. With OpenBuilds’ V-Slot profile parts, we fixed the elevator steps rigidly to each other with carriages and rollers that slide along the lifting mechanism. Ultimately, the carriages are moved by a belt stretched between a motor and tensioning unit mounted on the other side.

At the top of the last elevator step, we implemented a servo motor that unfolds a carbon fiber rod for watering flowers connected with a peristaltic pump installed in the water tank. Unlike standard rotary pumps, which are sensitive to the volume of liquid, we adopted peristaltic pumps, which squeeze an elastic tube through rollers on the circumference, and push the liquid out. Compared to standard pumps, these mechanisms have a much slower pumping speed but can lift liquid to a much greater height.

Technologies & tools

Back-end

Python, Django(DRF), FastAPI, AWS IoT Core, pandas, Loki, Prometheus, Grafana, API Gateway, AWS (Route, Lambda, RDS, S3, SQS, SES, EKS, ECR)

Front-end

JavaScript, TypeScript, React, Redux, Leaflet, Webpack, Axios, Material UI, Cube.js, AWS CloudFront

Embedded

AVR, Raspberry Pi, SPI, UART, USB, I2C, HTTP, Solidworks, ROS, SLAM, LiDAR, Altium Designer

ML/DS

OpenCV, TensorFlow, TFLite, ONNX, NumPy

DevOps

Terraform, Weave, Docker. Docker Compose, Kubernetes, BitBucket Pipelines

Database

PostgreSQL, AWS Timestream

Process

Our robotics department followed the agile methodology throughout the project, working closely with machine learning, computer vision, and data science specialists to achieve desired results. We strived to deliver a comprehensive solution without scope creep, demonstrating industry-specific knowledge to potential customers in a complex and demanding area. During regular meetings, brainstorming sessions, and retrospective analyses, our robotics experts kept up with the project’s progress and addressed all issues.

Currently, we test a watering and plant detection system and polish algorithm that automatically finds and reaches office plants at varying heights without colliding. We also identified design issues during development and built a sketch to address these side effects before showcasing the robot to investors. Also, our specialists develop a technical base for the robot, including a charging station connected to the water supply and 220V network, enabling the robot to charge the onboard battery and refill the built-in water tank automatically.

Team

Back-end Developers

Front-end Developers

Project Manager

Business Analyst

Software Architect

Team Lead

Hardware Engineers

Firmware Developers

DevOps Engineer

ML/DS Engineer

3D Modeller

Design Engineer

Results: 34% reduced plant damage with a smart IoT plant watering system

Innowise’s robotics team has built an IRIS – an automated IoT-driven robot to water plants and navigate office surroundings. We equipped the device with an advanced mapping system to build accurate routes through SLAM technology, LiDAR (laser scanners), and other sensors. Additionally, our engineers empowerуed the robot with an elevating mechanism based on sliding rollers and a carbon fibre rod on top.

As a result, we designed a watering system that allows the plants to be watered regularly without human interaction. IRIS ensures the flowers’ health, improving air quality and promoting a green atmosphere in the office. Furthermore, it reduces the workload of employees who previously had to water plants manually, allowing them to focus on their core responsibilities without being distracted by routine tasks.

Project duration

February 2023 - Ongoing

11%

savings on maintenance staff

34%

reduced plant damage

Related cases

Need a technological solution? Contact us!

What happens next?

Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.

After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.

We arrange a meeting with you to discuss the offer and come to an agreement.

We sign a contract and start working on your project as quickly as possible.

11% savings on maintenance staff after building an automatic plant watering robot

Customer

Challenge: Innowise's in-house project to showcase our robotics expertise