Case Studies
Berkshire Grey
3x Faster robotics dataset investigations: How Berkshire Grey drives productivity with FiftyOne
Jun 6, 2025
Berkshire Grey accelerated robotic vision model development and improved training data quality by adopting FiftyOne for multimodal data.
Sped up robotics data investigation and curation by
3x
Curation workflows resulted in datasets with robot picks
> 1M
Team collaboration features resulted in faster feedback loops improving
efficiency
In this article
Talk to a computer vision expert
Berkshire Grey is a leading robotics company delivering AI-powered robotic solutions to automate picking, sorting, and transferring tasks for optimizing material handling in warehouses. By utilizing accurate computer vision models that enable millions of robotics picks, Berkshire Grey has enabled leading retailers, e-commerce providers, and logistics companies such as Maersk and FedEx to achieve over 50% reduction in labor requirements while improving productivity and reducing errors.
“FiftyOne has really helped us speed up investigations. For example, if we see a wrong suction cup grasping an item, we can quickly visualize the issue across all data sources and identify what went wrong.” — Dimitry Pechyoni, Senior Principal Machine Learning Engineer at Berkshire Grey
Challenge: Fragmented data management and slow investigation speed hampered development
Before adopting FiftyOne, Berkshire Grey’s team relied on an internally developed system for visualizing and analyzing multimodal data, mostly images, video, and point clouds from their robotic systems. The data came from a variety of sensors, with hundreds of thousands of robotic pick actions being tracked daily. While this data was invaluable for troubleshooting and model training, the existing tool had several drawbacks:
- Slow issue investigation: The internal system lacked the visualization capabilities and interactivity for quick data exploration.
- Lack of interactivity: The internal tool only allowed viewing one example at a time, further limiting their ability to analyze issues effectively.
- Complex integration and customization: Adding visualizations for new sensors took months, and filtering based on multiple fields simultaneously was painfully slow.
Solution: Faster, more flexible data exploration with FiftyOne
Berkshire Grey discovered FiftyOne as the perfect fit for their data management and model investigation needs for their robotic systems. The platform’s ability to handle large, multimodal datasets with high performance and its user-friendly interface made it the perfect choice for their team’s needs.
FiftyOne’s interactive features allowed Berkshire Grey to visualize large datasets in a fraction of the time, with improved filtering and querying capabilities.
“The speed at which FiftyOne fetches data from our system is incredibly fast. When you're dealing with millions of data samples, that’s a game-changer. Most data management systems just can’t handle that scale.” — Dimitry Pechyoni, Senior Principal Machine Learning Engineer at Berkshire Grey
- Data management: The team developed custom data ingestion scripts, continuously importing data into FiftyOne, resulting in datasets with greater than 1M robot picks.
- Interactive filtering and tagging: The intuitive filtering system of FiftyOne allowed teams to slice data effectively, reducing troubleshooting time for identifying problematic robotic picks.
Enhanced collaboration: Tagging and URL-sharing features helped team collaboration, allowing members across multiple teams to quickly pinpoint and share interesting scenarios.
Results: 3x Faster investigations and data curation
After deploying FiftyOne, Berkshire Grey saw immediate improvements in their data workflows. New members were able to start using the tool’s features almost immediately after onboarding.
- 3x Faster investigations: The team was able to significantly reduce the time for issue investigation. What used to take more than a day could now be done in a fraction of the time. Filtering complex datasets became instant and interactive, boosting the team’s productivity and ability to diagnose problems.
“FiftyOne has really helped us speed up investigations. For example, if we see a wrong suction cup grasping an item, we can quickly visualize the issue across all data sources and identify what went wrong.” —Dimitry Pechyoni, Senior Principal Machine Learning Engineer at Berkshire Grey
- Improved data quality for model training: By using FiftyOne for data curation, the team was able to tag, filter, and eliminate poor-quality samples quickly, which improved the overall quality of the data used for training, leading to better model performance.
“Without FiftyOne, our investigation capabilities would be severely limited. Model training would grind to a halt as it entirely depends on the tool’s data management. FiftyOne is foundational to our robotics operations.”—Dimitry Pechyoni, Senior Principal Machine Learning Engineer at Berkshire Grey
Talk to a computer vision expert
Loading related posts...