本文へジャンプ

成果報告書詳細
管理番号20180000000151
タイトル*平成29年度中間年報 IoT推進のための横断技術開発プロジェクト ドメイン特化型IoTプラットフォームの研究開発
公開日2018/6/26
報告書年度2017 - 2017
委託先名国立大学法人東京大学 国立大学法人大阪大学 イーソル株式会社 株式会社アクセル
プロジェクト番号P16007
部署名IoT推進部
和文要約
英文要約Title: Project to develop cross-sectoral technologies for IoT promotion / Research and Development of Domain-Specific IoT Platforms (FY2017-FY2018) FY2017 Annual Report
(1) Development of compiler and hardware accelerators (The University of Tokyo)
We developed an LLVM-based compiler for a domain-specific heterogeneous System on a Chip (SoC) and implemented a backend for its ISA, called AXIS. Our compiler implemented almost all instructions and its test programs were implemented.
We developed object detection and position estimation accelerators for a domain-specific heterogeneous SoC with Axell Corporation. We selected Single Shot Multibox Detector (SSD) as an object detection accelerator and Normal Distributions Transform (NDT) as a 3D self-position estimation algorithm.
Towards hardware accelerators of SSD and NDT, we investigated the trade-off between the reduction of data (the number of coordinate bits) and the degradation of accuracy in an autonomous driving simulator.
(2) Development of heterogeneous SoC (Axell Corporation)
We presented a domain-specific heterogeneous SoC combining many-core and hardware accelerators, and completed architectural research and examination. Specifically, in order to assign the system task of the target domain to suitable computing resources, we investigated and analyzed the functions and performance required jointly with the University of Tokyo. Then we investigated and examined the architecture. We achieved the architectural research and examination 3 months ahead of schedule. We started consideration of specifications and implementation accordingly. Specifically, we examined the specification of the heterogeneous SoC, constructed a simulation environment, and developed a prototype compiler.
(3) Development of distributed processing platform for heterogeneous computing (Osaka University)
We evaluated the performance of existing many core processor environments. We used Kalray MPPA 256 and NVIDIA Drive PX 2 as existing many core environments.
On Kalray Access Core 2.8 we built socket communication processing and porting ROS and publish/subscribe operations between external ROS master and MPPA. We designed middleware based on publish/subscribe model as communication runtime.
(4) Development of high-performance, low energy consumption, and scalable OS (eSOL Corporation)
We extended an operating system for many-core processors and designed messaging APIs and shared/remote memory management. The operating system supports resource management mechanism and APIs for on-chip network. We verified the performance improvement under the communication between applications in on-chip network. In the field test, we performed an autonomous driving experiment with Autoware.
(5) Building evaluated platforms by autonomous driving vehicle (The University of Tokyo, Axell Corporation, Osaka University, and eSOL Corporation)
We built an existing autonomous driving vehicle using the NVDIA DRIVE PX2 as an evaluated platform and evaluated its performance in the field test. Our project including its demonstration was published in multiple media.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る