Research map / Reinforcement learning / Visual and representation RL

DEAL: Diffusion Evolution Adversarial Learning for Sim-to-Real Transfer

NeurIPS 2025 · Reinforcement learning › Visual and representation RL · approach: Skill learning and diffusion policies · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes DEAL, a sim-to-real transfer method for RL controllers that uses diffusion-based evolution with adversarial learning to bridge the reality gap (abstract truncated; details limited).

Problem

RL controllers trained in simulation often degrade significantly when deployed on real robots due to the reality gap; prior System Identification approaches improve simulator fidelity but face challenges such as collapse.

Approach

Uses diffusion evolution combined with adversarial learning to address sim-to-real transfer for RL controllers, as an alternative to purely improving simulator fidelity via system identification. Specific mechanism not stated in the available abstract.

Experimental setup

Baselines: System Identification

Builds on: System Identification

Closest papers by meaning

• EASI: Evolutionary Adversarial Simulator Identification for Sim-to-Real Transfer (NeurIPS 2024 · similarity 0.91)
• Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL (NeurIPS 2024 · similarity 0.86)
• Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
• ASID: Active Exploration for System Identification in Robotic Manipulation (ICLR 2024 · similarity 0.85)
• PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.84)
• Diffusion Policies Creating a Trust Region for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.84)