We are about to finish perception system of our autonomy stack. We have recently started working on planning and fusion for that we have been doing research on best available AI autonomous brain. And after few days we though of comparing all the available AI autonomous brains. We have come across many autonomous brain but these are the four which we really like. Of course it’s not possible to explain in detail about these autonomous brain but we have tried to cover brief so you can take a decision for your own personal application. If you know any autonomous brain which should be in this list please write in the comment, we will add.
Nvidia Drive PX2- AutoChauffer
NVIDIA DRIVE PX is the AI car computer that enables automakers, truck makers, tier 1 suppliers, and startups to accelerate production of automated and autonomous vehicles. It scales from a single processor configuration delivering Auto Cruise capabilities, to a combination of multiple processors and discrete GPUs designed to drive fully autonomous robotaxis.
The architecture is available in a variety of configurations. These range from one passively cooled mobile processor operating at 10 watts, to a multi-chip configuration with four high performance AI processors — delivering 320 trillion deep learning operations per second (TOPS) — that enable Level 5 autonomous driving.
The NVIDIA DRIVE PX platform combines
- Deep learning
- Sensor fusion
- Surround vision
It is capable of understanding in real-time what’s happening around the vehicle, precisely locating itself on an HD map, and planning a safe path forward.
It can fuse data from multiple cameras, as well as LiDAR, radar and ultrasonic sensors.
Direct connection of the vehicle harness cable to the harness connector, hence no need to change the connection of the car.
12- GMSL(Gigabyte Multimedia Serial Link) cameras can be connected(The GMSL Camera must be 8V tolerant)
DRIVE PX 2 contains network interfaces to provide the network connection for each Tegra chip.
Cost: $15,000
Specifications:
Architecture:
NVIDIA DRIVE PX 2 AutoChauffeur Documentation
NXP/Qualcomm — BlueBox
BlueBox is unique solution designed to help meet the stringent automotive safety, power and processing performance requirements of autonomous vehicle platforms supporting advanced classification tasks, object detection, localization, mapping and vehicle driving decisions.
The LS2 and S32V234 processors in the BlueBox run independent Linux systems, so the main host communication interface is the Linux terminal. Each terminal is accessible via UART or SSH over Ethernet. The recommended connection for most work is SSH over Ethernet, as it is a faster and more reliable connection than UART. However, to see the boot console for the LS2 or S32V234, you must use a UART connection.
BLBX2-DB BlueBox — Autonomous Driving kit
Open Platform
could be well-positioned to take advantage of a recent federal mandate to make automatic emergency braking a standard feature
BlueBox consists of two chips — one for processing visuals and a beefier one for making decisions. The bigger processor is eight integrated 64-bit ARM processors running at 2 GHz. NXP said the processor can run 90,000 million instructions per second at under 40 watts of power.
In addition to the processors, NXP provides much of computing power behind each of the sensors needed for autonomous driving — radar, vision and Lidar. The company also makes radio chips that lets vehicles talk to infrastructure — for example, a stop light that tells the car if it’s green, yellow or red.
NXP said four of the top five largest carmakers are already working on the BlueBox platform. Most of these cars will be released around the 2020 timeframe.
Enables adherence to ISO 26262 ASIL B standard for automotive safety applications
9.2 K DMIPS processing horsepower (without acceleration) for management of ADAS tasks
Enables automotive OS such as AUTOSAR to control interfaces to external devices without impacting Cortex-A53 performance
Security engine together with Arm TrustZone technology provides protection against IP theft and malicious hacking
Allows high-performance, low-power processing of incoming image data
Performs image housekeeping tasks such as HDR and color conversion, plus some dedicated image processing tasks
For rendering 3-D images; may also be used for additional image analysis tasks
Supports mono, stereo and surround view camera inputs; H.264 decode and encode also supported
DRAM support for LPDDR2/DDR3L/DDR3 for high-bandwidth data access, plus dual Quad SPI for external flash
EASE OF DEVELOPMENT
ROS Space
Open ROS Space Linux-based system
Programmable in linear C
Easily customizable
Development environment for mainstream vehicles
HIGHLY OPTIMIZED SENSOR FUSION
Various sensor data streams: radar, vision, LiDAR, V2X S32V234 automotive vision and sensor fusion processor
LS2084A embedded compute processor
S32R27 radar microcontroller
GPU: GC3000
Cost: $5,000
Architecture:
The production of Eyeq4 has not yet started so the datasheet is not available, the following information is taken from different sources.
Multiplethread Processing Cluster is used instead of GPU.
Memory: Compatible with LPDDR1,LPDDR2,DDR2,DDR3, SDRAM and also supports external flash.
Power consumption: 3 watts
Performance: 2.5 TFLOPS
EyeQ3 main feature which will be included in EyeQ4:
Four custom Vector Microcode Processors (VMPs)
800Mb/sec LPDDR2 SDRAM Controller
Serial Flash Interface (SFI) Serial Multi-SPI DDR/SDR FLASH/SRAM Controller
Arteris Interconnect
Three 16/12/8-bit 2Mpixel Video and Image preprocessing input ports
16-bit 2Mpixel Video output.
Two CAN ports (1Mbps)
Two UART ports (5Mbps)
Two I2C Interfaces (1Mbs)
Two SPI interfaces (31.25M (Master) / 15.62 (Slave) )
32-bit General Purpose I/O (GPIO)
Eight Timers (max output frequency 62.5MHz)
1Gb Ethernet / Audio Video Bridge (AVB)
Power management
Architecture:
Neuosys: Nuvo-5095gc
Nuvo-5095GC opens a new chapter for industrial computers. As the first embedded controller targeting at emerging applications of CUDA computing, autopilot, AI inference and virtual reality, Nuvo-5095GC integrates all features required for a compact, reliable and powerful GPU-computing platform.
Supporting nVidia GeForce GTX 950* and GTX 1050* GPU, Nuvo-5095GC possesses 768 CUDA cores to deliver tremendous computing power for arithmetic/graphics operations. Neousys’ patented Cassette technology and an innovative thermal design help to effectively dissipate the heat generated by GPU, thus make this compact system capable to operate reliably at 60°C with 100% GPU loading.
Nuvo-5095GC is based on Intel Skylake platform, and supports 35W/65W 6th-Gen Core™ processors and up to 32GB DDR4 memory. It offers rich I/O functions, such as GbE, USB 3.0 and COM ports, to connect external devices. All these extraordinary features are integrated into a very compact, 240 x 225 x 110 mm footprint. For fastgrowing GPU-computing applications, Nuvo-5095GC presents the first industrial-grade, compact and rugged platform incorporating CPU and GPU to offer performance far beyond traditional industrial computers.
Technical Specs:
Processor:
Supports Intel 6th-Gen Core LGA1151 CPU
Intel Core i7–6700 (8M Cache,3.4/4.0 GHz, 65W TDP)
Intel Core i5–6500 (6M Cache, 3.2/3.6 GHz, 65W TDP)
Intel Core i7–6700TE (8M Cache, 2.4/3.4 GHz, 35W TDP)
Intel Core i5–6500TE (6M Cache, 2.3/3.3 GHz, 35W TDP)
Temerature of operation: -25°C to 60°C
GPU: Independent nVidia GPU (75W TDP) or Integrated Intel® HD 530/510 Controller
Compatible with MezI interface for function expansion
Accommodates two 2.5” SATA HDD/SSD with RAID 0/1 support
I/O Interface:
Ethernet: 6x Gigabit Ethernet ports by Intel® 1x I219
USB: 4x USB 3.0 ports via native XHCI controller 4x and 4x USB 2.0 ports
Video Port
(Integrated Graphics ): 1x stacked VGA + DVI-D connector
2x DisplayPort connectors, supporting 4K2K resolution
Serial Port: 2x software-programmable RS-232/422/485 port (COM1 & COM3)
1x RS-232 port (COM2)
Audio: 1x Mic-in and 1x Speaker-out
Power consumption: 65 watts
Memory: Up to 32 GB, DDR4 and supports flash memory also
Price: $1,568
For More Information: https://pc-industriel.anteor.com
This research work has been done by Piyush R Medikeri. You can find the presentation video here
PS: Above all the information and details are based on our research and it is independent from any respective mention companies direct and indirect involvement.
