Research map / Reinforcement learning / Visual and representation RL
RL-CycleGAN: Reinforcement Learning Aware Simulation-to-Real
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add cycle consistency to GAN image adaptation | GAN · Robot 1 grasping | grasp success: 29% -> 61% | Helped |
| add RL-scene consistency to CycleGAN | CycleGAN · Robot 1 grasping | grasp success: 61% -> 70% | Helped |
| use grasping-specific masking loss for image adaptation | GAN-based simulation-to-real transfer · Robot 1 grasping | grasp success: 29% -> 63% | Helped |
| add RL-CycleGAN-adapted simulation to real off-policy training data | Robot 1 real-data RL training, 5,000 episodes · Robot 1 grasping | grasp success: 15% -> 75% | Helped |
| add RL-CycleGAN-adapted simulation to real off-policy training data | Robot 1 real-data RL training, 580,000 episodes · Robot 1 grasping | grasp success: 87% -> 94% | Helped |
| use RL-CycleGAN to adapt simulated images alongside real data | Robot 2 single-bin centered grasping, 3,000 episodes · Robot 2 single-bin centered grasping | grasp success: 13% -> 72% | Helped |
| use RL-CycleGAN to adapt simulated images alongside real data | Robot 2 single-bin centered grasping, 80,000 episodes · Robot 2 single-bin centered grasping | grasp success: 36% -> 95% | Helped |
| use RL-CycleGAN to adapt simulated images alongside real data | Robot 2 multi-bin randomized-location grasping, 80,000 episodes · Robot 2 multi-bin randomized-location grasping | grasp success: 33% -> 93% | Helped |
| fine-tune with RL-CycleGAN-adapted simulation and 5,000 off-policy episodes instead of RCAN's 28,000 on-policy episodes | Robot 1 grasping · Robot 1 grasping | grasp success: 90% -> 94% | Helped |
| set λRL-real to 0.1 rather than 1.0 (ablation) | RL-CycleGAN with 3,000 Robot 2 episodes · Robot 2 single-bin centered grasping | grasp success: 66% -> 72% | Helped |
| set λRL-real to 2.0 rather than 1.0 (ablation) | RL-CycleGAN with 80,000 Robot 2 episodes · Robot 2 single-bin centered grasping | grasp success: 91% -> 95% | Helped |
Closest papers by meaning
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.90)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL (NeurIPS 2024 · similarity 0.88)
- MENTOR: Mixture-of-Experts Network with Task-Oriented Perturbation for Visual Reinforcement Learning (ICML 2025 · similarity 0.88)
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
- RealVLG-R1: A Large-Scale Real-World Visual-Language Grounding Benchmark for Robotic Perception and Manipulation (CVPR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.