Store-Picker-Bot: A Major Step Forward in Real-World Robotics
We've made significant progress on Store-Picker-Bot as part of the Nebius x NVIDIA Global AI Hackathon 2026.
Our Raspberry Pi touchscreen dashboard is now connected to the physical robot over Wi-Fi, allowing us to interact with the existing ROS 2 and Nav2 system running on the Jetson Orin.
Here's what we've implemented and tested on the real robot:
- Live store map with LiDAR visualization and robot localization
- Autonomous goal navigation and multi-stop routes
- Saved shelf locations and map switching
- Automatic localization recovery when the robot's position estimate drifts
- Emergency STOP, communication watchdogs, and navigation safety checks
- A touchscreen interface for monitoring and controlling robot operations
One of our biggest challenges has been making navigation reliable on real hardware. We've been testing localization errors, sensor failures, and communication issues while improving the robot's recovery and safety mechanisms.
Next, we're working towards integrating NVIDIA Nemotron through Nebius Token Factory for AI-based task planning, followed by connecting the SO100 robotic arm for product picking and placement.
Our goal is to let a store operator submit an order in natural language and have the robot navigate to the required shelves, collect the products, and bring them to the packing area.
There's still work ahead, especially on the complete pick-and-place workflow, but getting the dashboard to control the real robot is an important milestone.
More updates coming soon as we continue building and testing Store-Picker-Bot.