platoseedplatoseed
Automated Scale AI for robotics data
Shotwell provides dense annotations and quality checks for robotics training data. Data quality is the biggest lever in improving robot model performance, and we have developed models that can automatically produce these annotations and operate at a fraction of the cost and instant turnaround time.
Shotwell AI provides a platform for labeling, collecting, and visualizing multimodal robotics data to help robotics teams build better models. It emphasizes fast, accurate, and dense annotations and action labeling across video data to improve training data quality.
The product offers tools to segment training videos into labeled actions, annotate robotics data frame-by-frame, and assess quality and task definitions. It supports labeling to create a dense, high-quality dataset for robotics training, with features to visualize and manage multimodal data and review labeled actions and SOP rubrics. The workflow includes ingesting videos, applying action annotations at a per-frame level, and reviewing labeled segments to support iterative model training.
Who it’s for: Robotics companies and teams focused on collecting and labeling multimodal data for robot perception and control models.
marketing site with demos and contact, mentions of trusted customers and open-source data demos; no explicit funding/hiring signals provided
Previously worked on humanoid robots at Sunday Robotics and self-driving cars at Tesla
Previously: co-founded Gradio which was acquired by Hugging Face and used by >1M developers a month, Tesla, MIT.
Currently working on unlocking robotics teams. Formerly @ Merge.dev, Roblox, Tarjimly (yc w18)
Documenting human dexterity
Multimodal data provider for robotics and world modeling
