Foundational science for
general cybernetics.
we believe that general-purpose robotics require general-purpose intelligence. iacon robotics open-sources core architectures, simulation algorithms, and safety bounds to accelerate the path to autonomous physical agency.
The Large Behavior Model: Generalizing Motor Control Across Morphologies via Causal Transformers.
abstract: traditional robotics relies on manually tuned state-space representations specific to a single morphology. we present the large behavior model (lbm), a causal transformer architecture trained on 14 billion synthetic physical state-transitions. by treating proprioceptive data and joint targets as a unified token stream, the lbm demonstrates zero-shot transfer capabilities between radically different kinematic structures.
0x7F2B_DEPLOY
SYNC_NOMINAL
| TITLE & SPECIFICATION | VENUE | CITATIONS | STATUS | ARCHIVE_REF |
|---|---|---|---|---|
01 Domain Randomization as a Primary Modality for Bipedal Transfer AUTHORS: E. Vance, M. Sterling, et al. | CoRL 2024 | 42 | PEER_REVIEWED | |
02 Unrolled Jacobian Solvers in High-Dimensional Contact Dynamics AUTHORS: A. Turing, J. Carmack | ICRA 2023 | 118 | PEER_REVIEWED | |
03 Safety Bounds for DRL on Heavy Machinery AUTHORS: M. Sterling, S. Connor | IROS 2023 | 205 | PEER_REVIEWED | |
04 Sparse Transformer Architectures for Latency-Critical Control AUTHORS: K. Dick, R. Zelazny | NeurIPS 2023 | 89 | PEER_REVIEWED | |
05 Multi-Agent Coordination in Adversarial Physics Envs AUTHORS: N. Stephenson, W. Gibson | CVPR 2024 | 34 | UNDER_REVIEW |
Open Datasets
academic access to industrial-scale trajectory data.
DEXTEROUS_GRASP_10M
petabytes of synthetic trajectory data generated by our hyper-parallel simulator. focused on multi-fingered manipulation.
BIPEDAL_RECOVERY_DB
14 billion state-transitions for bipedal robots under high-impulse external perturbations.
LBM_FOUNDATION_v2
pre-training weights and tokenized state-action pairs for the large behavior model v2 architecture.
Compute Grants
h100 cluster allocation.
we provide high-performance compute time to academic labs researching safe, generalized control policies and zero-shot sim-to-real transfer.
Partner with Iacon Research.
we are actively seeking collaboration with academic institutions and independent labs focusing on physical intelligence and autonomous motor control.