Research map / Reinforcement learning / Visual and representation RL
Visual Reinforcement Learning with Imagined Goals
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 |
|---|---|---|---|
| Use latent Euclidean distance rather than encoder-precision Mahalanobis distance for the reward (ablation) | RIG · Visual Pusher | distance to goal | Helped |
| Use latent-space distance rather than pixel MSE for the reward (ablation) | RIG · Visual Pusher | distance to goal | Helped |
| Relabel goals using a mixture of VAE-sampled and future goals (ablation) | RIG · Visual Pusher | distance to goal | Helped |
| Relabel goals using VAE samples only (ablation) | RIG · Visual Pusher | distance to goal | Helped |
| Relabel goals using future states only (ablation) | RIG · Visual Pusher | distance to goal | Helped |
| Remove goal relabeling (ablation) | RIG · Visual Pusher | distance to goal | Hurt |
| Train the VAE online on images collected during policy learning rather than leave it untrained (ablation) | RIG · Visual Pusher (easier goal space) | distance to goal | Helped |
| Fine-tune the VAE on exploration observations in addition to randomly generated observations (ablation) | RIG · Visual tasks | performance | Helped |
| Use RIG instead of prior image-based goal-conditioned RL methods | Visual Reacher, Visual Pusher, and Visual Multi-Object Pusher · Simulated visual manipulation tasks | distance to goal | Helped |
| Use RIG instead of state-based oracle RL | Visual Multi-Object Pusher · Simulated visual manipulation | distance to goal | Helped |
| Use RIG instead of state-based oracle RL | Visual Reacher and Visual Pusher · Simulated visual manipulation | distance to goal | Mixed |
| Train with two objects present, then test with varying object counts | RIG · Visual Multi-Object Pusher | performance | No clear effect |
Closest papers by meaning
- Learn Goal-Conditioned Policy with Intrinsic Motivation for Deep Reinforcement Learning (AAAI 2022 · similarity 0.90)
- Model-Based Visual Planning with Self-Supervised Functional Distances (ICLR 2021 · similarity 0.90)
- Unsupervised Control Through Non-Parametric Discriminative Rewards (ICLR 2019 · similarity 0.90)
- Discrete Compositional Representations as an Abstraction for Goal Conditioned Reinforcement Learning (NeurIPS 2022 · similarity 0.90)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.89)
- GoalLadder: Incremental Goal Discovery with Vision-Language Models (NeurIPS 2025 · 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.