Research map / Reinforcement learning / Policy gradient and actor-critic
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
Read the paperCheck 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 |
|---|---|---|---|
| switch DrQ's SAC backbone to DDPG (ablation) | DrQ · DMC hard tasks | exploration and task performance | Helped |
| add 3-step returns (ablation) | DDPG-based DrQ · DMC control tasks | sample efficiency | Helped |
| increase replay buffer capacity to 1M (ablation) | DrQ-v2 · Reacher Hard | performance | Helped |
| decay exploration noise linearly from 1.0 to 0.1 (ablation) | DrQ-v2 · DMC tasks, especially Humanoid | performance | Helped |
| add bilinear interpolation to random-shift augmentation (ablation) | DrQ-v2 · DMC tasks | performance | Helped |
| use mini-batch size 256 instead of 512 (ablation) | DrQ-v2 · DMC tasks | performance | No clear effect |
| lower the learning rate from 1×10⁻³ to 1×10⁻⁴ (ablation) | DrQ-v2 · DMC tasks | learning speed: 1×10⁻³ -> 1×10⁻⁴ | No clear effect |
| replace DrQ's random-crop augmentation with the custom grid_sample implementation (ablation) | DrQ · DMC image augmentation | training throughput: 2 | Helped |
| reimplement the replay buffer (ablation) | DrQ · DMC image-based tasks | storage capacity: ten-fold | Helped |
Closest papers by meaning
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.91)
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 2024 · similarity 0.90)
- Video-Enhanced Offline Reinforcement Learning: A Model-Based Approach (ICML 2025 · similarity 0.88)
- COVR: Collaborative Optimization of VLMs and RL Agent for Visual-Based Control (AAAI 2026 · similarity 0.88)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.88)
- Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control (NeurIPS 2024 · similarity 0.88)
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.