Flourish One: $3,555 Raspberry Pi Humanoid Built for Busy Parents

Flourish One is now available, and the company is promising an easy-to-use training interface. The startup, which operates from San Francisco and Paris, is opening sales for a $3,555 wheeled humanoid-style home robot. Early adopters can teach it new chores in under 30 minutes using only a smartphone.

Aiming at busy parents rather than factories, founder Antoine Marcel is betting that per-home, cloud-trained skills and phone-based teleoperation will turn the first batch of 50 robots into a proving ground for practical domestic robotics. Most humanoid and mobile-manipulator startups pursue factory or warehouse work, where environments are structured and data is plentiful. Flourish is deliberately taking on the messier challenge of the home.

Marcel’s thesis is that every household’s routines, layouts, and standards of cleanliness are too idiosyncratic for a one-size-fits-all generalist model. Flourish One therefore fine-tunes its skills task by task and home by home, using motion data captured from the owner’s phone and AI models running in the cloud.

If the approach works, it could become one of the first commercially available robots that busy families can actually train to handle real chores. A tiny team is hand-building the initial 50 units to learn what domestic robotics looks like outside the lab. The company was founded in January of this year, making it another humanoid company coming to market in less than 12 months.

Marcel is obsessed with time management. He told The Robot Report that he had automated everything on his computer, but when he came home, “the most painful stuff in my life was still there.” Asked why Flourish started with the home rather than an industrial application, he replied that every home is different—organization and routines are personal. Instead of waiting for a generalist AI model, the company fine-tunes per task and per home.

He explained that the way he tidies his apartment is not the same as someone else’s house. Users can show the robot for 30 minutes how to water plants, and Flourish will fine-tune an AI model for that task. The robot is then able to do it in that specific house, he said.

Flourish One has a wheeled base and two arms. The robot is built on a mobile base with six wheels for stability, and each of its two arms has a payload capacity of 1.5 kg (4 lb.). The battery is designed for 12 hours of arm operation.

Marcel noted that the company chose 1.5 kg because most tasks in homes do not require a huge payload. For obstacle detection, the robot uses a small lidar located on the base near the floor. Each two-fingered gripper also has a wrist-mounted camera for manipulation tasks.

This closeup of a prototype Flourish One shows the wrist camera orientation. Flourish One is not waterproof, so it will not do dishes or wash a dog in its first generation. The robot has limited dexterity but is suitable for tidying, taking out trash, and basic cleaning such as wiping a table, though not intricate assembly.

The robot is coming to market with a $3,555 price tag, and Flourish said it hopes to ship before Christmas. To help keep the bill of materials cost down, the unit is powered by a Raspberry Pi. The architecture will use cloud GPUs for training skill models and for inference of higher-level behaviors. It requires a network connection for “thinking,” or users can optionally run workloads on their own computers.

Marcel said he is optimistic that this architecture is the optimal solution for an affordable home robot. Putting an NVIDIA GPU in the robot would immediately double the cost, he said.

The company is developing a smartphone app that will use the inertial measurement units (IMUs) inside the phone for teaching and teleoperation. The user holds the phone and moves it as if performing the task, while the robot mirrors the action. According to Marcel, that is what is magical about Flourish. When training the robot to water plants, a user just takes the phone and moves it, and the robot will move its arms. No additional hardware is needed, he said.