System Requirements
LPU
The computing devices on which SENSR is running are call LiDAR Processing Unit (LPU). Seoul Robotics can supply its LPU-O, a specific Edge processing device that has been qualified for running SENSR from day 1. The LPU-O can process up to 8 LiDARs with 32 channels simultaneously, covering an area of about 15,000 m^2. For large sites, multiple LPU-O can be connected with each other to form a distributed system. For inquiries about Seoul Robotics' LPU, please reach out to our sales team at sales@seoulrobotics.org.
Requirement for your own computing device
SENSR can also be run in most machines, this section describes the requirements for running SENSR in a local machine.
GPU Consumption
SENSR detects vehicles through Deep Learning. This method provides greater accuracy but requires a higher capacity on the processing part. For optimal results it is highly recommended using a host equipped with a GPU.
GPU consumption is principally a function of the detection range: the bigger the detection range, the higher the consumption. The number of objects and points also affect it but in a secondary way.
A GPU Nvidia RTX3060 with 3,584 cuda cores can process an area of about 300x300 meters or 90,000 sqm. As a rule of thumb, the number of cores needed for a given site can be determined by
Nb Cores = Area to cover * 0.03982 .
CPU Consumption
CPU consumption is a function of:
- The LiDAR driver supplied by the manufacturer (some are more CPU-intensive)
- The number of points to be processed at the scene
- The number of concurrent objects at the scene
- Whether to use the LiDAR driver recovery and tilt auto-correction functions of SENSR
The following CPU examples are provided as a reference using 32 channels LiDARs with SENSR:
- For a single SENSOR: Intel Celeron J1900, Quad-core, 2.0GHz
- For up to 8 LiDARs: Intel 8th+ Gen i5 (i7 preferred) processor, >2.0Ghz
CPU threads can be allocated more flexibly in SENSR Algo settings > Performance. These parameters let the user choose the % of CPU you allocate to a given step on the pipeline.
The CPU consumption strongly depends on your device and its architecture. You can monitor the driver’s consumption using, for instance, htop and checking the processes that have the corresponding lidar manufacturer’s name (1 process per active driver/lidar).
RAM consumption
The RAM consumption is a function of:
- The number of points to be processed at the scene
- The number of concurrent objects at the scene
It is recommended using a machine equipped with at least 32Gb of RAM to ensure a smooth processing, 16Gb or RAM can work on sites counting 2 LiDARs and a detection range limited to 2500 sqm.
Disk footprint
The minimum requirement for the disk is 50Gb available to store the software images of SENSR, collect logs and operate smoothly.
Operating System
SENSR will only run on the following Operating Systems
- Ubuntu 20.04 LTS
- Ubuntu 22.04 LTS
- Redhat 8.X
- Jetpack 5.0 or above (for ARM devices)
If you need to run SENSR on a different OS, please contact Seoul Robotics.
GUI Mode with GPU Enabled
In both GUI and non-GUI modes, the GPU and CPU receive significant stress from SENSR. Non-GUI mode is slightly lower and therefore recommended in production.
Other Requirements
- SENSR has a GUI that uses OpenGL. The computer must have an OpenGL compatible graphics card.
- Make sure the /opt/ directory is read/write enabled. SENSR installs the project files and environment settings by default in the /opt/seoulrobotics folder.
In case of any doubt on the right hardware to use with SENSR, feel free to contact Seoul Robotics' support team.