This work develops an obstacle-avoiding tracking controller for a planar snake robot operating in a densely cluttered arena. We use NEAT to evolve both the structure and weights of a lightweight neural controller that converts local obstacle observations (simulated LiDAR) and robot state into two gait controls, frequency (speed) and offset (heading), within a serpenoid gait framework. A task-specific reward encourages forward progress, discourages corridor violations, and penalizes collisions. Experiments in PyBullet show effective, efficient navigation and a compact learned policy suitable for resource-constrained deployment.