Active Research Programs

EXCALIBUR is our flagship research program, conducted in collaboration with the U.S. Air Force and NVIDIA. The program develops next-generation perception and decision-making systems inspired by the architecture of the mammalian brain.

Our approach is grounded in predictive coding theory and the Free Energy Principle — frameworks from computational neuroscience that describe how biological systems build internal models of the world and act to minimize uncertainty. We implement these principles in real-time, embodied systems that process live sensory input on consumer hardware.

The system integrates multiple sensory modalities — vision and audition — using biologically-motivated architectures for routing, gating, and binding information across processing hierarchies. It has demonstrated emergent behaviors not explicitly programmed into the system, a key indicator of architectural fidelity to the underlying neuroscience.

We are developing novel learning architectures that more closely mirror how biological neural systems adapt over time. Unlike conventional deep learning — where a single optimization process runs indefinitely — our approach explores learning processes with variable lifespans, competing objectives, and state-dependent behavior.

This work draws on decades of research in synaptic plasticity, neuromodulation, and sleep-dependent memory consolidation to inform architectures where learning itself is dynamic, not static.

Our research builds on deep experience in GPU programming dating to NVIDIA's early CUDA platform. We develop and maintain custom computational pipelines for real-time neural simulation, large-scale model inference, and sensory processing — all running on local hardware with no cloud dependencies.

Our foundational work applies statistical mechanics to large-scale neural dynamics, building on a 12-year postdoctoral research collaboration at the Physical Studies Institute. This includes path integral formulations of neocortical interactions and the influence of macroscopic electromagnetic fields on synaptic processes.

These theoretical foundations — grounded in nonlinear stochastic differential equations and probability theory — inform the mathematical framework underlying our applied research programs.

Our research utilizes egocentric video, developmental recordings, and multimodal sensory datasets to train and validate perception models. We maintain active research partnerships with the U.S. Air Force Research Laboratory and NVIDIA.

We are committed to responsible use of human subjects data and adhere to institutional data governance policies for all research involving behavioral and developmental datasets. Researchers and institutions interested in collaboration are encouraged to contact us.