Robotics teams are collecting more data than ever, yet much of it never reaches training. In the search for “high-quality” samples, teams often discard large portions of what they collect.
Curation is painstakingly manual, and quality is rarely defined with enough rigor to be reproducible. Teams often fall back on hand-crafted heuristics, even when data comes from different operators, sessions, and conditions.
Hebbian Robotics is focused on bringing rigorous analysis to robotics datasets. We believe data should be studied with the same seriousness and methodology that researchers apply to models.
We come from top embodied AI and ECE programs at Columbia University and National University of Singapore. Our team has topped Stanford's LLM benchmarks, built infrastructure at Jane Street and Verkada, and is backed by Y Combinator and angel investors from Oracle and Google DeepMind.
Join us
If this thesis resonates with you, we'd love to hear from you. Explore Pareto and reach out to Brandon or Kingston.
Follow us on X at @hbr_pbc for updates.