The AI behind Brightpick

Intuition is our autonomous manipulation stack that fuses multiple modalities to give our robots the ability to perceive, reason, and act.

multi-modal ai

Intuition is a closed-loop system that learns and adapts with every pick

Vision

Visual intelligence

Each robot builds a detailed 3D point cloud of the tote and its contents. Depth sensing captures the shape and position of the items, while object segmentation separates individual products, even when they overlap or partially cover one another.

The AI identifies the optimal pick point for each item and selects the one with the highest probability of success.

Brightpick robots can also scan the destination container for gentle placing or pick-to-shipper, placing each item where it fits best to reduce damage risk and maximize space utilization.

touch

Tactile intelligence

As the gripper engages the item, pressure sensing enables the robot to precisely modulate downward force until a secure suction seal is established.

The robot monitors suction throughout the pick to ensure the item stays securely held until placement, with airflow adjusted for different items depending on weight, shape, and packaging.

Oftentimes robots carry automatic gripper exchangers, enabling them to select the best suction cup for each given item.

intelligence

Foundation model

Brightpick uses a two-stage AI approach to complete picks autonomously.

First, a lightweight 30M-parameter neural network runs directly on each robot, enabling fast, low-latency picking with over 98% success.

If the first attempt fails, the robot switches to a larger VLA model hosted on each warehouses’s on-site server, which combines a 3B-parameter backbone and 300M-parameter action expert and increases overall pick success to over 99.7%.

Human-in-the-loop

Exception handling

In the rare cases where an item remains too difficult to handle reliably, the robot makes up to three pick attempts before falling back to two possible human-assisted fallbacks:

Remote teleoperation: An operator remotely guides the robot through the pick.
Goods-to-person station: The tote is sent to a staffed workstation, where an operator completes the pick manually.

This ensures that, operationally, all orders can still be fulfilled without critical failure.

CLOSED-LOOP LEARNING

Every pick is a lesson

Each pick records what the robot saw, the action it chose, the result, and how any exception was handled. Successful picks, retries, failures, double-picks, and human interventions become training data.

Each manual intervention creates a valuable datapoint that helps train Intuition to handle similar edge cases autonomously in the future.

Because learning happens across hundreds of robots, each software update improves the entire fleet, increasing reliability and reducing cost per pick.

fleet management

Intelligent fleet orchestration

Brightpick Intuition orchestrates our entire robot fleet with precision, controlling every movement and fulfillment workflow across your warehouse.

At the core is our AI-powered Time-Space Planning algorithm, which continuously predicts fleet activity and dynamically re-plans robot paths to avoid congestion and maximize throughput.

Intuition also tracks the exact location, quantity, and batch of every SKU – giving you a live, accurate view of your inventory and the ability to instantly recall or cycle-count any item when needed.

ROBOTIC MANIPULATION HERITAGE

Brightpick’s picking history

Before Brightpick was founded in 2021, our team had already spent years working on robotic manipulation at Photoneo. That work saw the team deploy more than 1,000 robotic picking cells in industrial environments and factories.

This experience became the foundation for Brightpick’s mobile manipulators.

Instead of only moving products around the warehouse, Brightpick robots can identify items, pick them from real totes, handle edge cases, and keep order lines moving through the system.