Research map / Reinforcement learning / Offline RL
First-Person Activity Forecasting from Video with Online Inverse Reinforcement Learning
Nicholas Rhinehart, Kris Kitani
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 stop-based instead of scene-based goal discovery (ablation) | DARKO · Home 1 (action labels) | mean probability of true goal: 0.438 -> 0.614 | Improved |
| use stop-based instead of scene-based goal discovery (ablation) | DARKO · Office 2 (action labels) | mean probability of true goal: 0.238 -> 0.625 | Improved |
| use stop-based instead of scene-based goal discovery (ablation) | DARKO · Lab 1 (action labels) | mean probability of true goal: 0.426 -> 0.709 | Improved |
| replace visual detections with ground-truth labelled detections | DARKO · goal forecasting, all environments | mean probability of true goal: mean absolute improvement of 0.27 | Improved |
| add action features to state features in reward (ablation) | DARKO (labelled detections) · Home 1 | mean probability of true goal: 0.735 -> 0.851 | Improved |
| add action features to state features in reward (ablation) | DARKO (labelled detections) · Office 1 | mean probability of true goal: 0.581 -> 0.700 | Improved |
| add action features to state features in reward (ablation) | DARKO (labelled detections) · Lab 1 | mean probability of true goal: 0.892 -> 0.880 | Hurt |
| add state and action features over position-only reward (ablation) | DARKO (labelled detections) · Home 1 | mean probability of true goal: 0.674 -> 0.851 | Improved |
| add state and action features over position-only reward (ablation) | DARKO (labelled detections) · Lab 1 | mean probability of true goal: 0.886 -> 0.880 | Hurt |
| incorporate goal detection uncertainty into value iteration (ablation) | DARKO · synthetic noisy goal detections | average goal forecasting probability: p<10^-7 | Improved |
| use DARKO online IRL instead of RNN classifier | goal forecasting (visual detections) · Home 1 | mean probability of true goal: 0.291 -> 0.524 | Improved |
| use DARKO instead of Logistic regression | goal forecasting (labelled detections) · Home 1 | mean probability of true goal: 0.517 -> 0.851 | Improved |
Closest papers by meaning
- First-Person Activity Forecasting With Online Inverse Reinforcement Learning (ICCV 2017 · similarity 0.99)
- Visual Forecasting by Imitating Dynamics in Natural Sequences (ICCV 2017 · similarity 0.88)
- Predicting Goal-Directed Human Attention Using Inverse Reinforcement Learning (CVPR 2020 · similarity 0.87)
- Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos (NeurIPS 2022 · similarity 0.86)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.86)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · 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.
