Updates
- · latest · edited
RL Work#
Today, I returned to my reinforcement learning agent and developed several new variants building on DreamerV3.
The results mark another substantial leap forward. My strongest variant now achieves a normalized return of approximately 3.9%, up from roughly 3.0% previously, a relative improvement of about 30%. This brings my work to around 55% of the current state of the art, a considerable step closer to the goal.
Each iteration teaches me something new, and the momentum is unmistakable. Onward and upward!