Research map / Reinforcement learning / Visual and representation RL
Scaling All-Goals Updates in Reinforcement Learning Using Convolutional Neural Networks
Fabio Pardo, Vitaly Levdik, Petar Kormushev
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 a convolutional Q-map without compression instead of a goal-in-input network | Goal-conditioned pathfinding in random mazes · Random mazes, unseen test mazes | success rate: 99% | Improved |
| Use a Q-map without compression instead of a compressed Q-map (ablation) | Q-map maze pathfinding · Random mazes, unseen test mazes | success rate | Improved |
| Use Q-map all-goals outputs instead of goal-in-input outputs | Goal-conditioned pathfinding · Random mazes | success rate | Improved |
| Use Q-map all-goals outputs instead of goal-in-input outputs | Goal-conditioned box pushing · Sokoban, unseen 10×10 puzzles | success rate | Improved |
| Train the goal-in-input baseline with random goals rather than all goals (ablation) | Goal-in-input Sokoban baseline · Sokoban, unseen 10×10 puzzles | success rate | Mixed |
| Replace random actions with random goal-directed trajectories | Montezuma's Revenge exploration policy · Montezuma's Revenge | keys picked up: once; 398 times | Improved |
| Replace ε-greedy exploration actions with Q-map goal-directed trajectories | DQN task learner · Super Mario All-Stars | flags reached: 9 times; 33 times | Improved |
| Replace ε-greedy exploration actions with Q-map goal-directed trajectories | DQN task learner · Super Mario All-Stars | final performance: 30% higher | Improved |
| Transfer a pre-trained Q-map from another level instead of training from scratch (ablation) | Q-map goal-reaching model · Super Mario All-Stars, different level | learning speed | Improved |
Closest papers by meaning
- Adaptive Quasimetric Mapping : Principled Topological Abstraction for Robust Offline Goal-Conditioned Navigation (ICML 2026 · similarity 0.86)
- Learning World Models for Unconstrained Goal Navigation (NeurIPS 2024 · similarity 0.86)
- LAGMA: LAtent Goal-guided Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.85)
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.84)
- Exploring the Edges of Latent State Clusters for Goal-Conditioned Reinforcement Learning (NeurIPS 2024 · similarity 0.84)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
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.
