Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Shaping embodied agent behavior with activity-context priors from egocentric video
Tushar Nagarajan, Kristen Grauman
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 video-derived activity-context rewards | Ours · Cool, Store, Heat, Clean, Slice | success rate: Cool: 0.12 -> 0.26; Store: 0.00 -> 0.12; Heat: 0.01 -> 0.13; Clean: 0.35 -> 0.53; Slice: 0.30 -> 0.36 | Improved |
| Use video-derived activity-context rewards | Ours · Prep, Trash | success rate: Prep: 0.22 -> 0.26; Trash: 0.14 -> 0.13 | Mixed |
| Use uniform compatibility scores instead of activity-context scores (ablation) | Ours · AI2-iTHOR tasks | success rate: Cool: 0.07 vs 0.26; Store: 0.02 vs 0.12; Heat: 0.12 vs 0.13; Clean: 0.37 vs 0.53; Slice: 0.26 vs 0.36; Prep: 0.22 vs 0.2 | Hurt |
| Use word-embedding similarity instead of activity-context scores (ablation) | Ours · AI2-iTHOR tasks | success rate: Cool: 0.19 vs 0.26; Store: 0.02 vs 0.12; Heat: 0.03 vs 0.13; Clean: 0.40 vs 0.53; Slice: 0.34 vs 0.36; Prep: 0.22 vs 0.2 | Hurt |
| Use static-image spatial co-occurrence instead of activity-context scores (ablation) | Ours · AI2-iTHOR tasks | success rate: Cool: 0.33 vs 0.26; Store: 0.02 vs 0.12; Heat: 0.06 vs 0.13; Clean: 0.47 vs 0.53; Slice: 0.33 vs 0.36; Prep: 0.25 vs 0.2 | Mixed |
| Use interaction-label sequence transitions instead of detected activity-context scores (ablation) | Ours · AI2-iTHOR tasks | success rate: Cool: 0.33 vs 0.26; Store: 0.11 vs 0.12; Heat: 0.07 vs 0.13; Clean: 0.43 vs 0.53; Slice: 0.30 vs 0.36; Prep: 0.20 vs 0.2 | Mixed |
| Combine activity-context and navigation-exploration rewards | Ours · AI2-iTHOR tasks | success rate: Cool: 0.26 -> 0.25; Store: 0.12 -> 0.05; Heat: 0.13 -> 0.19; Clean: 0.53 -> 0.50; Slice: 0.36 -> 0.34; Prep: 0.26 -> 0.4 | Mixed |
| Replace ResNet-18 with ResNet-50 (ablation) | Ours · Cool, Store, Heat, Clean | success rate: Cool: 0.30 -> 0.15; Store: 0.16 -> 0.11; Heat: 0.11 -> 0.12; Clean: 0.55 -> 0.36 | Mixed |
| Replace LSTM with GRU (ablation) | Ours · Cool, Store, Heat, Clean | success rate: Cool: 0.30 -> 0.52; Store: 0.16 -> 0.31; Heat: 0.11 -> 0.17; Clean: 0.55 -> 0.73 | Improved |
Closest papers by meaning
- Embodied VideoAgent: Persistent Memory from Egocentric Videos and Embodied Sensors Enables Dynamic Scene Understanding (ICCV 2025 · similarity 0.88)
- Ego2Web: A Web Agent Benchmark Grounded in Egocentric Videos (CVPR 2026 · similarity 0.87)
- EgoExoLearn: A Dataset for Bridging Asynchronous Ego- and Exo-centric View of Procedural Activities in Real World (CVPR 2024 · similarity 0.86)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.86)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.86)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · 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.
