Hello, after reading your paper “Strengthening Generative Robot Policies through Predictive World Modeling,” I found your results truly amazing.
I would like to try out your framework myself, and I’m particularly curious about the computational cost involved. Specifically, when using the visual world model to evaluate and rank the generated action chunks, how much additional time does this process take? Also, for the alternative approach that performs direct gradient ascent optimization, how much computational time does that typically require?
Hello, after reading your paper “Strengthening Generative Robot Policies through Predictive World Modeling,” I found your results truly amazing.
I would like to try out your framework myself, and I’m particularly curious about the computational cost involved. Specifically, when using the visual world model to evaluate and rank the generated action chunks, how much additional time does this process take? Also, for the alternative approach that performs direct gradient ascent optimization, how much computational time does that typically require?