Till midterm evaluation: Hypothesized an actor-critic algorithm for deterministic fixed time free endpoint optimal control problem. This algorithm was thought of to generalize an algorithm (link) which has been observed empirically to have better convergence properties than policy gradient descent.

From midterm to final evaluation: Worked on getting empirical evidence for the proposed algorithm, but the algorithm failed to perform better than policy gradient approach. Found a theoretical explanation to the observations. The policy gradient approach is actually the actor-critic algorithm with noiseless critic.