C$^2$T: Captioning-Structure and LLM-Aligned Common-Sense Reward Learning for Traffic--Vehicle Coordination

C2T is a research-oriented project tackling the 'reward engineering' bottleneck in traffic management systems. While traditional Multi-Agent Reinforcement Learning (MARL) relies on narrow metrics like 'intersection pressure,' this project uses LLMs to interpret complex traffic scenarios and generate rewards that align with human common sense (safety, comfort, flow). Quantitatively, the project is in its infancy with 0 stars and 6 forks, indicating it is likely a recently published paper with initial interest confined to the academic peer group. Its defensibility is low because the core innovation—using LLMs for reward shaping—is a rapidly evolving technique popularized by projects like NVIDIA's Eureka. The primary moat would be the specific traffic-vehicle coordination dataset or the 'Captioning-Structure' logic, but as an open-source research implementation, it is easily replicable. Frontier labs like Waymo (Google) or NVIDIA are high-threat competitors as they already possess superior simulation environments and integrated hardware/software stacks for urban mobility. The 1-2 year displacement horizon reflects the high velocity of 'LLM-as-a-judge' and 'LLM-as-a-reward-function' research in RL.