Multi-Step Prediction for Curiosity Driven Learning

Ruchir Aggarwal*          Kushantha U. Attanayake*         Dennis Li*

                      Julio Soldevilla+               Poorani Ravindhiran^

For the results we present here:

• α = Scaler for loss in timestep t + 1 = 1
• β = Scaler for loss in timestep t + 2

Mario

• Comparison Video




• (Baseline) α = 1 , β = 0




• α = 1 , β = 0.2




• α = 1 , β = 0.5

Breakout

• α = 1 , β = 0.5

Space Invaders

• α = 1 , β = 0.5

Pacman

• Comparison Video




• (Baseline) α = 1 , β = 0




• α = 1 , β = 0.5