Research map / Reinforcement learning / Imitation learning
Learning from Trajectories via Subgoal Discovery
Sujoy Paul, Jeroen Vanbaar, Amit Roy-Chowdhury
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 |
|---|---|---|---|
| Discover ordered sub-goals from expert trajectories and use them to shape the reward | Subgoal discovery · AntMaze | cumulative reward: 0.0002 -> 0.012 | Improved |
| Add one-class out-of-set augmentation to the sub-goal reward (ablation) | Subgoal discovery · BiMGame | cumulative reward | Improved |
| Add one-class out-of-set augmentation to the sub-goal reward (ablation) | Subgoal discovery · AntTarget | cumulative reward | Improved |
| Use fewer sub-goals ($n_g=5$ rather than $n_g\geq 10$) (ablation) | Subgoal discovery · AntMaze | learning speed | Hurt |
| Use sub-optimal rather than optimal expert trajectories (ablation) | Subgoal discovery · BiMGame and AntMaze | learning performance | Hurt |
| Use sub-goal-based discrete rewards instead of value-function-based continuous rewards | Trajectory-guided RL · MuJoCo tasks | learning performance | Improved |
| Increase the number of sub-goals across a broad range (ablation) | Subgoal discovery · BiMGame and AntTarget | learning performance | No clear effect |
Closest papers by meaning
- Learning to Reach Goals via Iterated Supervised Learning (ICLR 2021 · similarity 0.91)
- Learning from Demonstrations via Capability-Aware Goal Sampling (NeurIPS 2025 · similarity 0.88)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.87)
- Identifying Selections for Unsupervised Subtask Discovery (NeurIPS 2024 · similarity 0.87)
- In-Trajectory Inverse Reinforcement Learning: Learn Incrementally Before an Ongoing Trajectory Terminates (NeurIPS 2024 · similarity 0.87)
- GO-DICE: Goal-Conditioned Option-Aware Offline Imitation Learning via Stationary Distribution Correction Estimation (AAAI 2024 · similarity 0.87)
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.
