We develop foundational infrastructure for robot learning—from data representation to cross-embodiment transfer. Our work enables the next generation of general-purpose robotic systems.
Unified formats for multi-modal robot data. Standardizing how observations, actions, and proprioception are encoded across heterogeneous platforms.
Learning policies that generalize across robot morphologies. Mapping action spaces between manipulators with different kinematics and dynamics.
Efficient imitation from human teleoperation. Extracting task structure and motion primitives from expert demonstrations at scale.
Integrating visual, proprioceptive, and tactile modalities. Temporal alignment and attention mechanisms for multi-sensor robot learning.
A comprehensive quantitative study of the simulation-to-reality gap using the Booster T1 humanoid robot. Analysis of 240 seconds of real-world data reveals leg joints exhibit 2× higher error than arms, with knee pitch joints showing 18.7° MAE. Provides actionable insights for domain randomization and sim2real transfer.
A standardized format for robot teleoperation data. Defines schemas for multi-modal observations, action spaces, sensor calibration, and coordinate frames. Designed for interoperability with LeRobot, RLDS, and Open X-Embodiment.
Formats and methods that work across robot types, sensor configurations, and task domains. No platform lock-in.
Reference implementations for every specification. If we propose it, we build it. Theory backed by working code.
Explicit semantics for every field. Coordinate frames, units, and conventions documented. No implicit assumptions.
We believe open standards accelerate the entire field. If you are working on robot learning infrastructure, data formats, or cross-embodiment transfer, we would like to hear from you.
RESEARCH@GERRA.COM