Research map / Label-efficient and robust learning / Knowledge distillation
Knowledge Distillation from Single-Task Teachers to Multi-Task Student for End-to-End Autonomous Driving
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a transformer-based RGB-D sensor fusion model for CARLA end-to-end driving that distills multi-task perception and uses an RL-based controller with PID waypoint following; the paper is a doctoral research proposal/plan rather than a completed evaluation.
Problem
End-to-end autonomous driving with RGB-D sensor fusion struggles in scenes with many dynamic agents, and imitation learning is limited by expert performance and out-of-distribution generalization.
Approach
Distill single-task teacher features into a multi-task transformer perception student (semantic segmentation, semantic depth cloud, ego speed, traffic light state), add 2D attention fusion and modified gradient normalization, and treat control as sequence modeling with an RL agent refined by PID waypoint following.
Key results
- No quantitative results are reported; the text states only that evaluation results demonstrated higher driving scores than prior work, and that final evaluation is planned for August 2024.
Limitations
- The text is a research plan with a timeline of planned work, so claimed components are not fully evaluated.
- No quantitative comparison numbers are given in the provided text.
Experimental setup
Datasets: CARLA simulator data
Benchmarks: CARLA
Metrics: route completion driving score
Baselines: LeTFuser EfficientNet-GRU baseline (Natan and Miura 2022a)
Builds on: LeTFuser TransFuser Decision Transformer Trajectory Transformer CvT
Closest papers by meaning
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.88)
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking (ICLR 2026 · similarity 0.88)
- SGDrive: Scene-to-Goal Hierarchical World Cognition for Autonomous Driving (CVPR 2026 · similarity 0.87)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.86)
- DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving (ICLR 2025 · similarity 0.86)
- DualAD: Disentangling the Dynamic and Static World for End-to-End Driving (CVPR 2024 · 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.