Simulated annealing solves CartPole without gradient-based optimization

The fact

This approach, based on random state jumps, bypasses limitations of traditional reinforcement learning methods.

The result opens perspectives for environments where gradients are difficult to compute or do not exist.