AI-driven robots may help in border management
Pravin Sawhney | Bengaluru
If things go according to plan, the Indian Army would have indigenous robots for border patrolling and early response in a decade. The Artificial Intelligence (AI) driven robots “cannot replace soldiers but would complement them by helping them in surveillance” says Chief Scientist, Manoj Jain, who heads the defence public sector undertaking Bharat Electronics Limited (BEL)’s Central Research Laboratory (CRL) at Bangalore.
The CRL has built technology demonstrators of first responder robots (meant for search and surveillance), and plan to have prototypes ready in one year, after which they could be given to the user (army) for trials and acceptance. Once done, they would be inducted into the service where they would work with and under human supervision. Robots, unlike humans, would not suffer from fatigue and would need little rest to deliver round the clock.
Giving credit to the secretary, defence production, Jain says that CRL is working in mission-mode to meet the users’ (all three services) requirements. After the defence ministry’s AI Task Force was set up, the services were asked to give their requirements. The army asked for use of AI in border management, gun management and electronic warfare management through its Director General Information System (DGIS). The air force evinced interest in image-based tracking and guidance in GPS denied environment, ground-based air defence system and so on, while the navy desired AI supported swarm technology and ships’ combat management system to name a few. According to Jain, all services are keen on AI based C4I (command, control, communication and intelligence) systems for good situational awareness. Before assessing how this would be done, a bit on AI itself would be in order.
AI is about simulating human intelligence in machines, especially computers, in three steps: learn, reason and self-correct. The computer is fed information (data) and rules for using information based upon which the machine on its own (reasons) comes up with approximate or definitive solutions by self-corrections. How good the final solution would be depends on five conditions which are essential for AI:
- The first is information or data, the more the better. As an analogy, what food is for humans, data is for machines;
- The second condition is computing power. The higher it is, more data it can process for accurate results;
- The third are human beings who can tag the data accurately into the computer;
- The fourth is the requirement for top computer scientists capable of writing algorithms from the tagged data on machines; and
- The fifth condition is more of a limitation. AI works in a single domain, not in cross domains. For example, a humanoid robot tasked for patrolling cannot do firing at the same time. This of course is no longer correct. With advances in AI, machine learning and deep learning neural networks (specifically by reinforced deep learning and transfer learning), cross domain tasks are possible, but this is outside the scope of this article.
Coming back to the CRL, Jain, who has been in this premier organisation for 28 years, told FORCE that the training from automation to AI-based solutions started in 2016. From mid-Nineties, the automation was introduced in two phases. Phase one, which lasted till 2000, involved digitisation of data and rules-based data analytics programmes and early-experts system. The second phase, though still automation, was premised on expert advice to the system (computer). “The whole idea was to shorten the sensor to shooter time with better accuracy through automation,” he says. Moreover, in all mission-critical systems (like ships’ combat management system), the expert advice (humans) oversaw and could overrule the system.
This is changing with the arrival of AI. While humans are still on the button, so to say, but the computer has enough leeway to learn, realise and self-correct. This big step was possible with the arrival of computers with exponentially improved computing power. Jain says, “We got super-computing or high-performance computing. General Purpose Graphics Processing Unit (GPGPU), which are used for core processing, give the speed of 1000 teraflop (close to 1 petaflop).”
AI is about simulating human intelligence in machines, especially computers, in three steps: learn, reason and self-correct. The computer is fed information (data) and rules for using information based upon which the machine on its own (reasons) comes up with approximate or definitive solutions by self-corrections
You must be logged in to view this content.