The implementation of AI coupled with hi-tech sensors offers massive potential for drones to become autonomous in the near future. Besides their implementation, the equipment and software that go into the UAVs will require even more enhancements. This development can benefit agriculture and disaster management and can lead India to play a greater role on a global scale.
Image credit (www.pixabay.com)
The huge benefits potential that drones provide for successful deployment in various sectors has resulted in growing interest from both public and private players. Thanks to emerging technologies such as AI and ML, these UAVs have a great scope for improvement. After witnessing the possibility of autonomous cars, a similar excitement has now surfaced around the possibility of having autonomous drones soon.
To discuss the latest technologies that can be leveraged to develop cutting-edge drones, a panel of technologists comprising of Daniel Raj David, CEO & Co-founder, Detect Technologies; Amit Mate, CEO & Founder, GMAC Intelligence; Sagar Gupta, CEO, Indian Robotics Solutions and Naresh Soni, CEO, Satvig, and moderated by Rahul Uniyal, Head of Aviation & Tech projects, Sterlite Power and Neel Mehta, Co-founder, Asteria Aerospace gave their expert opinions at the August edition of the Drone World 2020 webinar hosted by EFY.
Here’s presenting an in-depth look into how the entire conversation unfolded.
The topic of the discussion was ‘Latest technologies available to create cutting edge drones’ regarding which there were certain questions such as:
- What is the current state of technology for creating autonomous drones?
- What are the various challenges involving the use of a particular sensor?
- How critical it is to implement LiDAR?
- Can the upcoming technologies meet certifications for mass deployment and usage?
- How can drones help to counter the ongoing pandemic?
- What are the benefits that early adopters of drones can expect to have?
- How effective is the integration of AI/ML?
- What is important: cost or application?
- What are exciting tech use cases for drones in the next 5 years?
Issues to address
Before providing answers to the above questions, the challenges that worry most developers of drone technology were identified such as :
- Uncertainty with respect to the operation of a drone at any given time, especially in the dark.
- Ultrasonic sensors can sometimes give inaccurate data while landing a drone and can lead to instability of the UAV.
- Difficulty in inspection and positioning of non-GPS terrains like inside steel/petroleum plants.
- Dependency of LiDAR on GPS. The moment there is no GPS, LiDAR fails to work.
- Scepticism of a client around using new tech.
- Constraints due to government norms and regulations.
- Current drones face memory, compute and storage limitations.
- An AI/ML model running on cloud can’t help.
- Very expensive to run face recognition using cloud API
- Ultrasonic sensors and LiDAR can be used during low visibility operation for obstacle avoidance.
- For determining a proper height while landing, monocular cameras can be installed onto the drones.
- LiDARs have a good range and can give direct measurement (achieve up to 5 cm accuracy) without ground control.
- LiDAR can provide better quality imagery data where the RGB mapping system fails.
- LiDARs can be used for mapping power lines, greenfield surveys, smart cities and mining.
- Because of mass-scale production, the cost of LiDAR has come down.
- Simultaneous Localisation and Mapping (SLAM) technology can be implemented for underground mapping in the absence of GPS.
- To tackle unprecedented situations (for example created by COVID-19), face recognition technology or number plate recognition technology can be used.
- As it is cumbersome to deploy autonomous machines with heavy electronics in a place like a hospital, usage of AI/ML instead can help in offering an integrated solution in terms of hardware.
- Integration of AI/ML into Beyond Visual Line Of Sight (BVLOS) can help to avoid accidents.
- Edge can enable various use cases less expensively.
- Concerning cost, people need not worry as there are solutions for all price points. At the same time, go for scalable computing platforms.
- By installing a thermal camera on a drone with good flight time, then the required data can be obtained.
”When you don’t correlate the imagery data (obtained through photochromatory process), LiDAR can then give better quality data. These are not exactly opposite technologies but complement each other. When it comes to cost, most people think that LiDARs are expensive. In the last 4-5 years because of autonomous cars and the emergence of new technologies, mass-scale production is happening resulting in cost reduction of LiDARs,” said Naresh Soni.
Work to be done
- Commercial usage of SLAM technology needs lots of research.
- For developing autonomous drones, the safety of people should be a priority.
- Core hardware innovations need to be built into the drones involving positioning of electronic chips at a certain area for localisation purposes.
- More research regarding usage of wideband frequencies for positioning of a drone.
- Research regarding drones that can come in contact and take accurate measurements.
- Lots of testing and investment required for developing technologies that can be implemented in difficult conditions.
- AI/ML layers and models need to be understood well for the integration of various sensors while developing autonomous drones.
- Since LiDAR provides density values and produces grayscale imaging, colour can be added to data using RGB mapping.
”In autonomous, we need to ensure about safety..it is the first thing because there are several obstructions when flying in a city such as narrow street or trees. A drone might require to be operated both at day and night. Therefore obstacle avoidance sensor is necessary to be used in each and every drone, said Sagar Gupta.
”Drones are vehicles with eyes. If you position chips in certain areas, then it acts as a localisation sensor but the core research there ends up being what sort of technology it will use (ultra-wideband, ultrasound etc.). On the hardware front, the question regarding drones to go in contact with structures stably using positioning systems needs to be attended,” said Daniel Raj David.
Expectations for the future
- Drones should become more compact and powerful.
- Besides being autonomous, drones should incorporate a variety of sensors and be capable of operating regardless of the availability or unavailability of GPS.
- Drones should be employed more for delivering goods and medicines during disaster management.
- Develop drones that can come in contact to perform various functions that involve complex dynamics.
- Focus should be more towards the agriculture sector.
- Developers must collaborate and compete at an international level.
”What is going to drive this industry over the next decade or so is a metric called picojoule per inference. It relates to the amount of power consumption to compute one AI/ML inference on an edge device. It applies to make the drones autonomous as well as generate revenue streams for people who want to deploy these drones,” said Amit Mate.
”Our competition is at a different plane. It is first with the international players. Lastly, the drone has to (implemented) as a service. The hardware will reduce to less than 5 per cent of the value of the drone but 90 – 97 per cent of value will come from the software that is packed into it. Its future value depends on what you do with the data, not the collection of data,” said Amber Dubey.
- By using some of the latest technologies available such as LiDAR, drone technology can bring a lot of value to use cases.
- AI/ML involving face recognition can help create phenomenal solutions for many industries.
- Lastly, the core development process should be centred towards making in India with an eye on competing with the world players. If the drones work in India, then they can work anywhere in the world.