日本フィジカルAI新聞

世界のフィジカルAIを、日本語で。

End-to-End自動運転arXiv:2609.21404

SC-IMMベースの教師信号によるEnd-to-End自動運転の軌道出力安定化

Stabilizing Trajectory Outputs in End-to-End Autonomous Driving via SC-IMM Based Teacher Signals

シェア:XThreadsFacebookLINEはてブBluesky

シーン条件付きIMMで専門家軌道を経路-速度状態に変換し、補助教師信号として学習に加えることで、End-to-End自動運転モデルの閉ループ走行安定性を向上させた研究。

詳しい要約

1. どんなもの?

- End-to-End autonomous driving models predict future waypoints from sensor inputs and convert them to control commands. - Conventional waypoint-based imitation learning minimizes coordinate-level errors, failing to capture scene-dependent path-speed changes and temporal instability. - Proposes an offline teacher-signal generation and learning method for trajectory-output stabilization based on Scene-Conditioned Interacting Multiple Model (SC-IMM). - Converts expert trajectories into path-speed…

2. 先行研究と比べてどこがすごい?

- Conventional waypoint-based imitation learning mainly minimizes coordinate-level errors, making it difficult to capture scene-dependent path-speed changes and temporal instability across waypoint outputs. - The proposed method addresses these limitations by incorporating scene-conditioned motion-model cues as auxiliary supervision. - In closed-loop evaluation on 100 short routes in CARLA Town12, the method improved driving score by 28.0% and reduced Collision/km by 62.3% compared with the base…

3. 技術・手法の肝は?

- Converts expert trajectories into path-speed states. - Performs IMM updates conditioned on scene cues to generate path-speed teacher labels and mode posterior probabilities. - Adds generated signals to the original trajectory loss as auxiliary supervision during training. - Inference structure and waypoint controller remain unchanged.

4. どうやって有効だと検証した?

- Closed-loop evaluation on 100 short routes in CARLA Town12. - Compared with baseline, driving score improved by 28.0% and Collision/km reduced by 62.3%. - Also improved jerk and trajectory-variation metrics.

5. 議論はある?

- The results demonstrate that offline teacher signals embedding scene-conditioned motion-model cues can guide trajectory-output driving models toward more stable closed-loop behavior. - No explicit limitations or potential negative aspects are discussed in the abstract.

6. 次に読むべき論文は?

- The abstract does not reference specific prior works or related methods. - Next papers to read could include foundational works on End-to-End autonomous driving, waypoint-based imitation learning, Interacting Multiple Model (IMM) filters, and closed-loop evaluation in CARLA.

※ AIが要旨から生成した要約です。正確性は原文をご確認ください。

著者: Siewoo Kim, Seung-Hyun Kong

分類: cs.RO

原文アブストラクト

End-to-End autonomous driving models commonly predict future waypoints from sensor inputs and convert them into vehicle control commands through a downstream controller. However, conventional waypoint-based imitation learning mainly minimizes coordinate-level errors, making it difficult to capture scene-dependent path-speed changes and temporal instability across waypoint outputs. In this paper, we propose an offline teacher-signal generation and learning method for trajectory-output stabilization based on a Scene-Conditioned Interacting Multiple Model (SC-IMM) to mitigate this issue. The proposed method converts expert trajectories into path-speed states and performs IMM updates conditioned on scene cues to generate path-speed teacher labels and mode posterior probabilities. The generated signals are added to the original trajectory loss as auxiliary supervision during training, while the inference structure and waypoint controller remain unchanged. In closed-loop evaluation on 100 short routes in CARLA Town12, the proposed method improved the driving score by 28.0% and reduced Collision/km by 62.3% compared with the baseline, while also improving jerk and trajectory-variation metrics. These results demonstrate that offline teacher signals embedding scene-conditioned motion-model cues can guide trajectory-output driving models toward more stable closed-loop behavior.

関連論文

  1. 1Boston Dynamics、車輪付きAtlasの研究試作機を紹介研究・技術
  2. 2GXOとApptronik、Apolloの物流現場試験を拡大 最大1,000台の展開を構想量産・展開
  3. 3newmoとKDDIが資本業務提携 2028年に堺市で自動運転タクシー商用化目指す提携・買収
  4. 4NVIDIA、CoRLでGR00T 2とIsaac Lab 3.0を発表新製品・発表
  5. 5UC BerkeleyとNVIDIAの研究者、AIエージェントでロボット制御コードを仮想環境で改善研究・技術

PR・アフィリエイト広告

FRONTIER(GPU搭載BTOパソコン)

FRONTIER(GPU搭載BTOパソコン)の公式広告バナーFRONTIERでGPU搭載PCを見る ↗
PR本紙発行元 EmplifAI