Bhopal-Based Swaayatt Robots Demonstrates New Obstacle Avoidance Patterns

Sanjeev Sharma, founder and CEO of Bhopal-based AI and robotics startup Swaayatt Robots, recently shared a video in which he showcased the company’s autonomous vehicle negotiating traffic cones (obstacles) at very aggressive speeds. 

The cones’ locations were not pre-programmed, which required the vehicle to compute its trajectory and make impromptu decisions. To further test the framework’s capabilities, the system was constrained to act on obstacles only within a 24-metre radius.

This innovation aims to extend the company’s previous claim of achieving Level-5 autonomy, in which vehicles can operate in all environments without human intervention. To put it in perspective, Tesla cars fall under Level 2 of the six levels of vehicle automation. 

“Our vehicle, arriving at the intersection, slows down for the speed breaker and then decides which lane or gate to commit to,” Sharma told AIM in an interview last year while noting that this capability fell under Level 5 autonomy.

New Demonstration

The demonstration included two obstacle avoidance patterns: zig-zag manoeuvres and lane-specific obstacle navigation. For the latter, the system brought the vehicle back to the left lane at aggressive speeds after avoiding obstacles. 

The Bolero vehicle used in the test, known for its high body roll, consistently maintained speeds above 45 kmph during zig-zag navigation and above 50 kmph during aggressive planning phases, slowing only slightly at critical moments.

“A human driver would face difficulties when driving beyond 40 kmph in such scenarios,” Sharma added.

The algorithmic framework, running on a single-threaded i7 processor at over 800 Hz, includes five classical agents and one reinforcement learning agent. 

It features a trajectory tracking algorithm that generates adaptive commands for the pure-pursuit algorithm to calculate the steering angle, with a single hyper-parameter controlling aggressive behaviour.

What’s Next?

The framework will be further scaled using end-to-end deep reinforcement learning, with additional demonstrations planned for February or early March. This advancement marks a significant step toward safer, fully autonomous vehicles.

Swaayatt Robots has been developing novel motion planning and decision-making frameworks designed to enhance the performance of autonomous vehicles. 

This new development allows vehicles to navigate smoothly and at high speeds while avoiding obstacles in real time.

Last year, Anand Mahindra, chairman of Mahindra Group, praised the company for trying to achieve Level 5 autonomy. “Sanjeev is using complex math to target Level 5 autonomy. I’m cheering loudly. And certainly won’t debate his choice of car,” he had said. 

To accelerate its growth, the company also raised $4 million at a valuation of $151 million from US investors in June last year. “By the end of this year (2024), we’ll be creating a blueprint that could solve Level Four Autonomy globally. To scale that model, we may raise around $1.5 billion,” Sharma told AIM last year.