Voxel51 blog

Physical AI is AI that perceives and acts in the physical world: robots, autonomous vehicles, and drones. Posts cover curating and evaluating sensor data in FiftyOne, from MCAP logs to VLA policies.

Visual AI is how machines learn to see: detection, segmentation, embeddings, and vision-language models. These posts show how to debug, compare, and improve visual AI models with FiftyOne.

Search is how you find what matters in datasets too large to review by hand. FiftyOne makes visual data searchable by natural language, similarity, and metadata to surface edge cases and rare scenarios.

Data curation is choosing what to train on: surfacing edge cases, removing duplicates and leaky splits, and closing coverage gaps. These posts show how to do it at scale in FiftyOne.

Label quality caps model performance. These posts cover auto-labeling with foundation models, agentic labeling with the Voxel51 Agent, and finding label errors before they reach training.

A single accuracy number hides more than it reveals. Slice performance by class and scenario, analyze failure modes, and compare detection, segmentation, and VLM models side by side in FiftyOne.

Every model is a portrait of its dataset. We load open datasets like COCO, LIBERO, and Open X-Embodiment into FiftyOne to reveal the imbalances, errors, and hidden gems summary stats miss.

Plugins make FiftyOne fully extensible. Build custom panels, operators, and workflows, or connect annotation tools, vector databases, and model APIs. If FiftyOne doesn't do it yet, a plugin probably can.

What's new at Voxel51: FiftyOne open source and Voxel51 releases, partnerships, ML research, and events like CVPR. Start here to see where the platform for physical AI is headed.