Research map / Multi-agent systems and games / Human-AI and goal-conditioned
AGIL: Learning Attention from Human for Visuomotor Tasks
Ruohan Zhang, Zhuode Liu, Luxin Zhang, Jake A. Whritner, Karl S. Muller, Mary M. Hayhoe, Dana H. Ballard
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 predicted-gaze attention mask channel to the policy network | imitation policy network · MsPacman (Atari) | action prediction accuracy (%): 54.6 -> 70.6 | Improved |
| add predicted-gaze attention mask channel to the policy network | imitation policy network · Seaquest (Atari) | action prediction accuracy (%): 46.0 -> 61.6 | Improved |
| add predicted-gaze attention mask channel to the policy network | imitation policy network · Breakout (Atari) | action prediction accuracy (%): 81.5 -> 86.2 | Improved |
| add predicted-gaze attention mask channel to the policy network | imitation policy network · Freeway and Riverraid (Atari) | action prediction accuracy (%): 96.7 -> 96.4; 72.5 -> 72.5 | No clear effect |
| add foveated image channel built from ground-truth gaze | imitation policy network · MsPacman (Atari) | action prediction accuracy (%): 54.6 -> 65.1 | Improved |
| use attention mask instead of foveated rendering | policy network · Seaquest (Atari) | action prediction accuracy (%): 51.4 -> 61.6 | Improved |
| add AGIL attention to plain imitation learning | imitation policy · Breakout (Atari game score) | game score: 1.6 -> 19.9 | Improved |
| add AGIL attention to plain imitation learning | imitation policy · Seaquest (Atari game score) | game score: 252.2 -> 788.9 | Improved |
| add AGIL attention to plain imitation learning | imitation policy · Freeway (Atari game score) | game score: 29.6 -> 30.6 | No clear effect |
| add motion (optical flow) channel to the image-only gaze network (ablation) | gaze network · Venture (Atari) | NSS: 5.429 -> 6.222 | Improved |
| add bottom-up saliency channel to the image-only gaze network (ablation) | gaze network · eight Atari games | NSS, AUC, KL, CC: Image 6.336 vs I+S 6.363 NSS on Breakout | No clear effect |
| add motion channel to image-only gaze network (ablation) | gaze network · eight Atari games | NSS, AUC, KL, CC: Image 5.776 -> I+M 5.834 NSS on Seaquest | Improved |
Closest papers by meaning
- Gaze-VLM: Bridging Gaze and VLMs through Attention Regularization for Egocentric Understanding (NeurIPS 2025 · similarity 0.85)
- Inverse Attention Agents for Multi-Agent Systems (ICLR 2025 · similarity 0.85)
- Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.84)
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.84)
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · similarity 0.84)
- Selective Visual Representations Improve Convergence and Generalization for Embodied AI (ICLR 2024 · similarity 0.84)
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.
