Research map / Reinforcement learning / Visual and representation RL
Generating Long-term Trajectories Using Deep Hierarchical Networks
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 |
|---|---|---|---|
| add attention to hierarchical policy (vs concatenation H-GRU-CNN-CC) (ablation) | HPN · Basketball holdout Delta=0 | micro-action accuracy: 31.5 -> 33.7 | Helped |
| add hierarchy with attention vs flat GRU-CNN | GRU-CNN · Basketball holdout Delta=0 | micro-action accuracy: 25.8 -> 33.7 | Helped |
| add memory (GRU) to CNN micro-policy (ablation) | CNN · Basketball holdout Delta=0 | micro-action accuracy: 21.8 -> 25.8 | Helped |
| feed-forward stack of look-ahead predictions instead of attention (ablation) | HPN · Basketball holdout Delta=0 | micro-action accuracy: 33.7 -> 26.9 | Hurt |
| also train on weak attention labels phi (ablation) | H-GRU-CNN-ATT · Basketball holdout Delta=3 | micro-action accuracy: 30.5 -> 28.0 | Hurt |
| also train on weak attention labels phi (ablation) | H-GRU-CNN-ATT · On-policy rollouts | rollout quality | Helped |
| model dependency between look-ahead predictions (ablation) | HPN · Benchmark | accuracy | No clear effect |
| HPN vs CNN baseline in expert preference | CNN · Preference study, experts | average gain: 0.68 | Helped |
| HPN vs ground truth in expert preference | Ground truth · Preference study, experts | average gain: -0.12 | No clear effect |
| HPN vs hierarchical baseline without attention | H-GRU-CNN-CC · Preference study, experts | average gain: 0.76 | Helped |
| non-expert judges instead of experts, vs GRU-CNN | HPN · Preference study | average gain: 0.68 -> 0.52 | Mixed |
Closest papers by meaning
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 2024 · similarity 0.86)
- Data Augmentation for Instruction Following Policies via Trajectory Segmentation (AAAI 2025 · similarity 0.86)
- Latent Plan Transformer for Trajectory Abstraction: Planning as Latent Space Inference (NeurIPS 2024 · similarity 0.85)
- PcLast: Discovering Plannable Continuous Latent States (ICML 2024 · similarity 0.85)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.85)
- Sports-Traj: A Unified Trajectory Generation Model for Multi-Agent Movement in Sports (ICLR 2025 · similarity 0.85)
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.