As machine learning enthusiasts, we’ve all been there – limited by budget and safety concerns when working with real robots. That’s why simulation platforms can be a lifesaver. But have you ever tried using pi0, the well-known VLA model, in simulation? I’d love to hear about your experiences.
I’m currently facing the same constraints, and I’m looking for alternative ways to test and train my models. pi0 seems like a promising solution, but I’m concerned about distribution shift. Would it be a significant issue in simulation?
Sharing your insights would be incredibly helpful. Have you encountered any challenges or successes with pi0 in simulation?