PILLAR-Robots in Agrifood

AI2Life is an important partner in the EU project PILLAR-Robots

The agrifood sector is undergoing a profound transformation, driven by the need to increase productivity, reduce waste, and ensure consistent quality standards. Within this context, PILLAR-Robots introduces adaptive, purpose-driven robotic solutions designed to operate in dynamic post-harvest environments.

Scenario 1: Removing Rotten Fruits and Vegetables

In a typical post-harvest setting, different types of fruits/vegetables—both fresh and rotten—are stored in various containers, such as plastic cassettes or smaller paper/plastic trays. The task to remove rotten fruits/vegetables from ripe and unripe may be simple in appearance but complex in execution.

In the task “Throw away rotten bananas”, the PILLAR-Robot must distinguish among different levels of ripeness, identify the item to pick, and discard it. This requires advanced perception capabilities, including visual recognition of spoilage patterns and colour variations. Through purposeful learning, the robot continuously improves its ability to distinguish between good and defective items, helping reduce food waste, improve hygiene, and support workers in repetitive or potentially unpleasant tasks.

Scenario 2: Picking Fruits and Vegetables to collect orders

In another common use case, different types of fruits/vegetables are placed on a table or in plastic cassettes. The task now is goal-oriented and quantity-based: “Pick 2 bananas.” Here, the PILLAR-Robot learns to:

  • Recognise specific fruit types
  • Count the requested quantity
  • Pick up the selected items
  • Place them carefully into designated crates
The mobile manipulator TIAGo robot involved in the pick and collect scenario with fruits and vegetables

This capability supports flexible packaging operations, especially in small and medium-sized enterprises where product combinations may change frequently. The robot adapts to new requests without requiring extensive reprogramming, increasing operational agility.

Agrifood Scenario with Pre-Conditions

To study the adaptation capabilities of Purposeful robots more deeply, we designed a scenario in which the robot can interact with a scene that enables more complex behaviours, such as sequential ones.

Scenario 1: Removing Rotten Fruits and Vegetables with Pre-Conditions

In this enhanced scenario:

  • Different fruits and vegetables (rotten and good) are present.
  • Fruits and vegetables are always located in a specific openable container/box

The task is “Throw away rotten tomatoes.” The PILLAR-Robot must not only identify and remove spoiled tomatoes but also:

  • Understand that relevant fruits and vegetables are confined to a specific container,
  • Recognise the specific container and vegetables/fruits also if they are not visible and blocked in a box
  • Explore the environment to find new interactions with objects, like the button that allows the opening of the box,
  • Pick up the items of interest,
  • Place them carefully into designated crates.

Scenario 2: Picking Fruits and Vegetables with Pre-Conditions to collect orders

In this enhanced scenario:

  • Mixed fruits and vegetables are available,
  • Fruits and vegetables are always in a designated container.

The task is “Pick 2 bananas and 2 apples.” The PILLAR-Robot must not only identify and pick the specific fruit/vegetables, but also:

  • Understand that relevant fruits and vegetables are confined to a specific container,
  • Recognise the specific container and vegetables/fruits also if they are not visible and blocked in a box,
  • Explore the environment to find new interactions with objects, like the button that allows the opening of the box,
  • Pick up the items of interest,
  • Place them carefully into designated crates.

Towards Adaptive and Human-Centred Automation

Across these scenarios, PILLAR-Robots demonstrates how purposeful, intrinsically motivated learning can enhance automation in the agrifood sector. By combining perception, manipulation, and contextual reasoning, the robot supports workers, increases efficiency, and helps meet higher quality-control standards.

These use cases illustrate a step toward more flexible, intelligent, and human-centred robotic systems—capable not only of executing predefined tasks, but of learning, adapting, and operating reliably in complex real-world environments.

PILLAR-Robots