Infoprop-Dyna: Uncertainty-Aware MBRL
- Infoprop-Dyna is an uncertainty-aware model-based reinforcement learning framework that builds a world model from real data to generate synthetic rollouts for effective policy optimization.
- It explicitly separates aleatoric from epistemic uncertainty and uses information propagation to terminate rollouts before accumulating excessive predictive error.
- Deployed on the Mini Wheelbot, it enables rapid skill acquisition and superior racing performance, achieving reliable laps in just 11 minutes of real-world learning.
99 99^ 99 99 OR \999 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99) OR \999 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 OR \999 99 99 99 99 99^ 99 99 OR \999 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99^ 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99^ 99 99 99 99 99 99 99 99 99 99 99^ 99 Dyna model-based reinforcement learning (Frauenknecht et al., 28 Jan 2025)999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99^ 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 99 99 99 99 99 99 99 OR \999 99 99 99 99 99^ 99 OR \999 OR \999 99^ 99 99 99 99 99 99^ 99 OR \999 99^ 99 99 99 99 99 99 99 99 99 99 OR \999 99 99 99 99 99 99 99 OR \999 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 Dyna model-based reinforcement learning (Frauenknecht et al., 28 Jan 2025)999 99 OR \999 99 99 OR \999 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 Dyna model-based reinforcement learning (Frauenknecht et al., 28 Jan 2025)999 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 Dyna model-based reinforcement learning (Frauenknecht et al., 28 Jan 2025)999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 OR \999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 OR \999 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 OR \999 99 99 OR \999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 Dyna model-based reinforcement learning (Frauenknecht et al., 28 Jan 2025)999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99) OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 OR \999 99 99 99 99 OR \999 99 99 99 99 99 99 99 99 99) OR \999 OR \999 99^ 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99 99