IWR6843: MMwave vs ultrasonic proximity sensors for mobile robots

Jens Lindrup

Part Number: IWR6843

Tool/software:

I am working with an outdoor mobile robot, with an operating speed at maximum 2 m/s. For collision avoidance on the robot, I am considering mmwave radar or ultrasonic sensors. Could you help with advice on choosing the correct sensor type going forward. I am especially concerned if signal processing and tuning for mmwave will be very extensive.

Best regards

Jens

4 months ago

0 Pedrhom Nafisi 4 months ago

TI__Genius 17877 points

Hello Jens,

For robotics, TI recommends the IWR6843 mmWave Radar Sensor. We have the Radar Toolbox that includes free source code for a bunch of different applications such as People Tracking, liquid Level Sensing, and Robotics. These demos include the signal processing to generate a point cloud and other types of output, and more processing on the point cloud to get information such as object position and human/non-human classification.

https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AJy2dB7pfZ0nJqp9U0xVSg__radar_toolbox__1AslXXD__2.20.00.05

https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AOWrkwbTC6SqhjS8xzVlrg__radar_toolbox__1AslXXD__2.20.00.05

For Robotics we have a safety bubble demo and a small obstacle detection demo. They use ROS and we also provide a standalone ROS Driver for outputting the point cloud which can then be combined with any other ROS compatible sensor of interest.

https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AFIEVaaBoBCgo.VCtJenZQ__radar_toolbox__1AslXXD__2.20.00.05

https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AFBXnwiM2XV37eEtzWKflA__radar_toolbox__1AslXXD__2.20.00.05

Comparing difficulty of signal processing and tuning between radar and ultrasonic, both are going to be more complex than light based sensors such as a Camera, LiDAR, or ToF sensors. Light based sensors work like our eyes do, so it is generally easier to understand, but we have collateral showing how radar data compares to what we see in our Radar Academy.

https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AEIJm0rwIeU.2P1OBWwlaA__RADAR-ACADEMY__GwxShWe__LATEST

Best Regards,

Pedrhom

0 Jens Lindrup 4 months ago in reply to Pedrhom Nafisi

Prodigy 10 points

Dear Pedrhom,

Thank you for your answer, we have ordered a IWR6843ISK radar test kit. Would it be possible to run both small obstacle detection and the safety bubble with the same radars on our robot?

Also, we are experiencing a lot of artifacts and noise with our radar - does it require a lot of tuning to work? We are a bit concerned about robustness when deploying on our robots.

Best,

Jens

0 Jens Lindrup 4 months ago in reply to Jens Lindrup

Prodigy 10 points

Sorry, we have ordered a IWR6843AOP EVM. Can I still run the small obstacle detection and safety bubble demos with this kit?

0 Pedrhom Nafisi 4 months ago in reply to Jens Lindrup

TI__Genius 17877 points

Hello,

6843ISK and 6843AOP are fundamentally the same chip, the difference with AOP is that the antenna is on the package itself and the antenna geometry being different. This requires one change in the configuration file to adhere to this difference but will still allow compatibility with all 6843ISK Demos.

There needs some tuning, but ultimately it is not a device/technology that needs to have a different configuration for every slight change in environment, that would not be feasible. We have tuning guides across the Radar Toolbox and E2E forums for different devices and demos.

The biggest factor you will notice is a mentioning of CFAR threshold. Think of it as a filter that only allows detections of a certain reflectivity level or higher to be shown in the point cloud.

Best Regards,

Pedrhom

0 Jens Lindrup 4 months ago in reply to Pedrhom Nafisi

Prodigy 10 points

Hello,

Thank you very much! So is 6843ISK also an evaluation board, or something we can use in production?

I read somewhere that the antenna geometry of the ISK is suited for wall mount applications and AOP is ceiling mount. Is that correct?

Could you please specify the one change in the config file so I can use this demo with the AOP board: https://dev.ti.com/tirex/explore/node?a=VLyFKFf__4.11.0&node=A__ALTzjndbFvzim8cyGM-VSQ__radar_toolbox__1AslXXD__LATEST ? That would be very helpful.

Best regards

Jens

0 Jens Lindrup 4 months ago in reply to Jens Lindrup

Prodigy 10 points

Hello again,

I also need to know, whether this sensor is best suited for outdoor operation? According to this document: https://www.ti.com/lit/ab/swra758/swra758.pdf?ts=1725598822336&ref_url=https%253A%252F%252Fwww.google.com%252F it states "The IWR6843 mmWave radar device can be used to detect people in a variety of indoor settings"

0 Pedrhom Nafisi 4 months ago in reply to Jens Lindrup

TI__Genius 17877 points

Hello,

The 6843ISK is an evaluation module, the chip itself being the 6843 or more specifically IWR6843, however we have reference designs which include the antenna as seen on the EVM itself. The 6843 is TI's highest performance non-automotive mmWave radar sensor, so it will be the best at pushing max detection ranges.

Best Regards,

Pedrhom

0 Jens Lindrup 4 months ago in reply to Pedrhom Nafisi

Prodigy 10 points

Thank you!

I would like to try the Mobile tracker demo on my robot, can you help me modify the IWR6843ISK chirp file for this?

Can I use IWR6843AOP on a platform driving at 2 m/s?

0 Pedrhom Nafisi 4 months ago in reply to Jens Lindrup

TI__Genius 17877 points

Hello,

For AOP I would recommend starting with 3D People Tracking example which has a prebuilt AOP image prepared. The Mobile Tracker and Small Obstacle Detection demos are literally the 3D People Tracking example with an extra occupancy zone feature and nothing else (and some different configurations). You can still use the 3DPT example for a moving platform and you will get experience on how radar detects objects when a platform is stationary vs dynamic

Best Regards.

0 Jens Lindrup 4 months ago in reply to Pedrhom Nafisi

Prodigy 10 points

Thank you Pedrhom. I have been investigating the 3D people tracker and it seems to have very good performance when the radar is stationary, but poor performance when the radar is moving.

The objective of this task is to create a system where I detect any object and slow down if it is within a specific FOV. Maybe I can utilize the safety bubble demo for this to get started?

I am a bit stuck in where to begin. Thanks.

+1 Pedrhom Nafisi 3 months ago in reply to Jens Lindrup

TI__Genius 17877 points

Hello,

mmWave Radar can definitely do the application well and we have plenty of footage of doing so:

https://www.ti.com/video/6207205419001

An extensive guide on the People Tracking algorithm chain and tuning it can be found below

C:/ti/<RADAR_TOOLBOX_DIR>/source/ti/examples/People_Tracking/docs/3D_people_tracking_detection_layer_tuning_guide.pdf

For the purposes of configuration, you do not need to know the details behind the algorithm implementation, but knowing what parameters you can adjust or knobs you can turn and what these adjustments affect.

dynamicRACfarCfg's <rangeThre> and <angleThre> which is what we call CFAR thresholding is what impacts the amount of points you see the most, as this dictates how many points get filtered out based on a point's range and angle reflectivity level.

Best Regards,

Pedrhom

Sensors

Sensors forum

IWR6843: MMwave vs ultrasonic proximity sensors for mobile robots