Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Coaching a Teachable Student
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 |
|---|---|---|---|
| Use CaT with aligned BEV features, feature distillation, safety hints, and student-paced coaching | RGB driving agent · Longest6 | Driving Score (DS): 48.41% to 58.36% | Helped |
| Add BEV-based safety hints to the student-teacher setup (ablation) | CaT · Longest6 | Driving Score (DS): 49.69% to 55.55% | Helped |
| Add IPM-based image-to-BEV alignment (ablation) | CaT student · Longest6 | Driving Score (DS): 39.48% to 44.10% | Helped |
| Add three-layer feature distillation rather than distilling one layer (ablation) | CaT student · Longest6 | Driving Score (DS): 45.23% to 55.55% | Helped |
| Add segmentation and command auxiliary supervision (ablation) | CaT student · Longest6 | Driving Score (DS): 48.07% to 53.23% | Helped |
| Add entity attention to the teacher BEV (ablation) | Privileged BEV teacher · Longest6 | Driving Score (DS): 65.73% to 73.30% | Helped |
| Add agent forecast channels to the teacher BEV (ablation) | Privileged BEV teacher · Longest6 | Driving Score (DS): 52.81% to 65.73% | Helped |
| Use full multi-term feature distillation instead of direct three-layer L2 matching (ablation) | CaT student · Longest6 | Driving Score (DS): 49.31 to 55.55 | Helped |
| Add Chamfer Distance to three-layer L2 feature distillation (ablation) | CaT student · Longest6 | Driving Score (DS): 49.31 to 51.95 | Helped |
| Use student-paced coaching instead of periodic coaching (ablation) | CaT student · Longest6 | Driving Score (DS): 48.97 to 58.36 | Helped |
| Emphasize hard samples with hard mining rather than student-paced coaching (ablation) | CaT student · Longest6 | Driving Score (DS): 44.92% vs. 55.55% | Hurt |
| Set initial coaching weight lambda_0 to 0.8 instead of 0.5 (ablation) | Student-paced coaching · Longest6 | Driving Score (DS): 58.36 to 51.76 | Hurt |
Closest papers by meaning
- Student-Informed Teacher Training (ICLR 2025 · similarity 0.89)
- Feedback-Guided Autonomous Driving (CVPR 2024 · similarity 0.86)
- Knowledge Distillation from Single-Task Teachers to Multi-Task Student for End-to-End Autonomous Driving (AAAI 2024 · similarity 0.86)
- Multi-Teacher Knowledge Distillation with Reinforcement Learning for Visual Recognition (AAAI 2025 · similarity 0.85)
- Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents (NeurIPS 2026 · similarity 0.85)
- Mimicking the Maestro: Exploring the Efficacy of a Virtual AI Teacher in Fine Motor Skill Acquisition (AAAI 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.