Estimate, Don't Imitate: Reusing Differentiable State-Based Policies for Visuomotor Control | TickerVault