# AI models for picking and placing fresh produce

The picking and placing AI models of TOS Robotics tell a robot where to pick each product and where to place it. They are trained from demonstrations with the Teachbot and see each product as a 3D point cloud, so they handle products that all differ. You buy them as a licence for robots of any brand. Terms on request.

Source: https://tosrobotics.com/products/picking-and-placing-ai-models/
Last updated: 2026-11-20

- Pick accuracy: 99.1% in production
- Trained from: Demonstrations with the Teachbot
- Sees: 3D point clouds
- How you buy it: Licence, terms on request

Every bell pepper has its own shape. **A robot that expects a fixed position misses.** Ours looks at each product and decides how to pick it.

## How one pick works

1. **3D cameras look at the product.** They give the shape, the size and the position of each product.

2. **The AI model decides the grip.** For each product it gives the robot a grip point and an approach angle.

3. **The robot picks on the fly.** The robot follows the product, whether the belt moves or stands still.

4. **The product is placed.** In the tray, the crate or the flowpacker infeed, in the position the pack needs.

## How you buy it

The models are sold as a licence for robots of any brand. Terms on request. They run in our [AI robot control software](https://tosrobotics.com/products/ai-robot-control-software/) and in every cell of TOS Robotics. How they are built is explained under [AI algorithms](https://tosrobotics.com/capabilities/ai-algorithms/).

## Performance

- picks per minute, per robot, on the bell pepper line: 40
- pick accuracy, up from 97.6% in the first weeks after go-live: 99.1%
- robot cycles in production: 2.3M+

## Where it is used

- [Bell pepper flowpacking with a singulator and AI robots](https://tosrobotics.com/cases/bell-pepper-flowpacking-singulator/): Robots fill the flowpacker, where people used to stand.
- [Mango packing in trays and crates with robots](https://tosrobotics.com/cases/mango-flowpacking/): Ready for deployment. Mangos into carton trays of 2, or into small and normal crates.
- [Avocado packing in trays and crates with robots](https://tosrobotics.com/cases/avocado-flowpacking/): Ready for deployment. Avocados into carton trays of 2 or 4, or into crates.
- [Assembling meal packages with robots](https://tosrobotics.com/cases/meal-packages/): Ready for deployment. A modular cell that fills meal boxes per recipe, without hands at the line.

## Keep reading

- [AI algorithms: imitation learning and 3D vision](https://tosrobotics.com/capabilities/ai-algorithms/): Imitation learning, a diffusion policy on 3D point clouds and clean data from the line.
- [How to automate flowpacking of fresh produce](https://tosrobotics.com/knowledge/automate-flowpacking-fresh-produce/): The flowpacker wraps on its own, the handwork is in feeding it, and TOS Robotics automates that in three steps.

## Questions and answers

### Can a robot pick products that all have a different shape?

Yes. The AI model of TOS Robotics does not need a fixed shape or position. It looks at each product with a 3D camera and decides per product where to grip and how to place it. That is why it works on bell peppers, avocados and mangos, which are never identical.

### Which robots do the models work with?

Robots of any brand. The bell pepper line runs on FANUC robot arms. The TOS Robot, our own arm for high-speed picking and placing, uses the same models.

### How does the model learn a new product or pack?

TOS Robotics teaches it by demonstration with the [Teachbot](https://tosrobotics.com/products/teachbot/). The AI model learns from those examples. No pick positions are programmed.
