Development of Lifelong SLAM System for Service Robots using Deep Semantic Information
| Project/Area Number |
21K14115
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 20020:Robotics and intelligent system-related
|
|---|
| Research Institution | Tohoku University |
|---|
Principal Investigator |
Ravankar Ankit 東北大学, 工学研究科, 特任准教授 (40778528)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Project Status |
Granted (Fiscal Year 2022)
|
| Budget Amount *help |
¥4,030,000 (Direct Cost: ¥3,100,000、Indirect Cost: ¥930,000)
Fiscal Year 2023: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2022: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2021: ¥2,340,000 (Direct Cost: ¥1,800,000、Indirect Cost: ¥540,000)
|
|---|
| Keywords | SLAM / Navigation / Deep Learning / Semantic Understanding / Path Planning / Robotics / Mobile Robots / Semantic Mapping / Scene Understanding / 自律ロボティクス / 人工知能 / HCI |
|---|
| Outline of Research at the Start |
本研究では「シーン外の経験」を導入することで新しい課題を処理できるセマンティックSLAMシステムを構築することを目的とする。このスキームによってロボットは新しい環境で効率的に学習することが可能となる。本研究は、認知科学の分野で議論されている「セマンティックシーンの理解」と「物体認識」の知見にインスピレーションを得ており、Deep Reinforcement Learning 構造を利用することによって現在のAIに足りない「少ないデータ」での「フレキシブル」な学習の実現を目指す。
|
| Outline of Annual Research Achievements |
This study aims to develop a lifelong semantic navigation system for mobile robots in dynamic environments by addressing two main problems, 1. Semantic understanding of everyday objects in indoor environments for long-term mapping and navigation, and 2. Planning under uncertainty for object search, detection, and exploration using multi-layered semantic maps. Data collection in simulated and real environments and tests for search-based optimized planning has been conducted. The project aims to enhance human-robot interaction in service robots by providing deep understanding of daily objects and their associations with locations, and developing an ontology for lifelong learning. This system can improve performance in tasks like search and exploration in dynamic settings.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
In the fiscal year, we implemented and tested the proposed framework in both simulated and real environments, using a house-like setting for data collection and testing various mapping and navigation algorithms. We completed the multi-layered semantic map pipeline and heuristic search and planning in diverse scenarios. Sensor integration on the robot system and deep learning training on common household objects were accomplished. The project resulted in a new multi-layer semantic mapping framework with metric maps, object maps, and search region maps within the simulated house.
|
| Strategy for Future Research Activity |
In the next fiscal year, we plan to conduct a long-term mapping and navigation study within Tohoku University's living lab facility (a simulated house). We aim to enhance robot planning by adding new layers to the semantic map to account for human presence including layers for privacy and understanding social interactions for navigation. We also plan to extend the framework to incorporate multi-robots for improved object search in semantic maps and develop task allocation algorithms. Integration with IoT devices for speech-based interaction is also on the agenda. Lastly, we will test cutting-edge deep learning networks to optimize the power and performance efficiency of the navigation system.
|
Report
(2 results)
Research Products
(62 results)
