Research map / Reinforcement learning / Visual and representation RL
Addressing Sample Complexity in Visual Tasks Using HER and Hallucinatory GANs
Himanshu Sahni, Toby Buckley, Pieter Abbeel, Ilya Kuzovkin
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 |
|---|---|---|---|
| Add HALGAN goal hallucinations to HER replay | DDQN · TurtleBot pebble collection | learning onset | Improved |
| Add HALGAN goal hallucinations to HER replay | DDPG · MiniWorld Navigate | task completion | Improved |
| Add HALGAN goal hallucinations to HER replay | DDPG · MiniWorld Pick&Place | task completion | Improved |
| Use naive HER without visual goal hallucination | DDQN · TurtleBot pebble collection | learning onset: 100K steps | Improved |
| Use naive HER without visual goal hallucination | DDPG · MiniWorld Navigate | task completion | Hurt |
| Use naive HER without visual goal hallucination | DDPG · MiniWorld Pick&Place | task completion | Hurt |
| Use VAE-based reward reassignment with HER instead of hallucinating goals | DDPG · MiniWorld Navigate | task completion | Hurt |
| Use VAE-based reward reassignment with HER instead of hallucinating goals | DDPG · MiniWorld Pick&Place | task completion | Hurt |
| Use dense rewards from random goal images | RIG- · MiniWorld Navigate | task completion | Mixed |
| Use dense rewards from random goal images | RIG- · MiniWorld Pick&Place | task completion | Mixed |
| Reduce HALGAN training snapshots to 1000 images (ablation) | HALGAN+HER · TurtleBot pebble collection | learning speed: 1000 images | Hurt |
| Add an L2 penalty to the generator output to encourage minimal image edits (ablation) | HALGAN · HALGAN goal hallucination | hallucination minimality | Improved |
Closest papers by meaning
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 2024 · similarity 0.87)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.87)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.87)
- HASARD: A Benchmark for Vision-Based Safe Reinforcement Learning in Embodied Agents (ICLR 2025 · similarity 0.86)
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.86)
- Cluster-Based Sampling in Hindsight Experience Replay for Robotic Tasks (Student Abstract) (AAAI 2024 · similarity 0.86)
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.
