Technical Intelligence Corpus

A research framework (EvoDriveVLA) for improving Vision-Language-Action (VLA) autonomous driving models by evolving training/distillation through collaborative perception–planning, using self-anchored perceptual constraints and future-informed trajectory optimization to reduce perception drift and long-horizon planning instability after visual encoder unfreezing.

A principled, efficient Transformer model (PAINET) for 3D dynamics modeling in multi-body systems, targeting trajectory/interaction prediction beyond what is directly observed.

Extrusion-focused segmentation strategy to improve CAD reconstruction quality from point clouds (reverse engineering / quality control), as described in an accompanying arXiv paper.

Provide a portable “trust network” for applications, people, and AI agents (identity/trust relationship layer).

Multimodal, speech-based conversational assistants that can tool-call to control/interact with smart-home/IoT devices, incorporating speech understanding plus modeling of IoT device state and spatiotemporal constraints with mixed-initiative interaction.