VladoiuGabriel/ped-traj-intent-pred

The project is a very early-stage (21 days old) implementation of a Vision-Language-Action (VLA) approach for a specific autonomous driving sub-task (pedestrian trajectory). With 0 stars and 0 forks, it currently lacks any community traction or ecosystem moat. While the VLA approach is a 'hot' research topic in the wake of models like RT-2 and Wayve's GAIA-1, this specific repository appears to be a personal experiment or a student project applying known VLM/VLA patterns to the nuScenes dataset. Defensibility is minimal as the core logic is likely a wrapper around standard Transformer/Vision backbones. The 'Frontier Risk' is high because massive players in the autonomous vehicle (AV) space (Waymo, Tesla, Wayve) and foundation model labs (Google DeepMind, OpenAI) are aggressively developing end-to-end world models and planning agents that would render such a niche tool obsolete. There is no evidence of a proprietary dataset or novel architectural breakthrough that would prevent a larger lab from replicating or surpassing this performance as a side effect of training more generalized driving models.