Discovery Channel to showcase Hungarian scientific research
Discovery Channel, the international factual television channel takes viewers into the world of self-driving cars and autonomous drones in one of its popular series, Made In with Csaba Gyetván.
In the latest episode, the programme visits the Systems and Control Laboratory (SCL) of the HUN-REN Institute for Computer Science and Control (SZTAKI).
Hungarian research on Discovery Channel
Airing on 28 December at 7 PM, the episode introduces fully autonomous parcel delivery drones operating without any central control, and also explores whether Csaba Gyetván can defeat a model car whose control algorithm has undergone 3,000 hours of learning.
The episode presents a system developed by the laboratory that is capable of planning drones’ autonomous flights in advance and responding to unexpected situations in real time. The drone independently collects a parcel from a moving vehicle and then delivers it with centimetre-level precision to another vehicle that is also in motion.
First in the world
To carry out this task, the system is provided only with the vehicles’ starting and end positions; all other calculations and control decisions are handled by algorithms developed by SZTAKI SCL.
At present, this complex operation can be performed fully autonomously and without human intervention only by experts at the Hungarian laboratory, making it unique worldwide.
Viewers will also see how the same system can simultaneously manage the flight of up to twenty drones, some of which are virtual. Instead of relying on instructions from a central controller, the drones plan and optimise their routes by using trajectory data from one another.
They are not only aware of where obstacles are at any given moment, but are also able to predict how those obstacles may move in the future. Based on these predictions, the drones adjust their own paths while maintaining safe distances at all times.
The research can help finding solutions to real life problems
The laboratory is also home to the development and testing of artificial intelligence-based control algorithms that enable autonomous vehicles to better predict, react and learn in complex road environments. Researchers simulate and model scenarios that would be too dangerous or even impossible to test in real-life conditions.
Models that have already been validated in theory and through computer simulations are then tested using virtual reality (VR), augmented reality (AR) and mixed reality (MR) technologies.
Engineers can create virtual hazards such as a child stepping onto the road from behind a parked bus, a cyclist making a sudden manoeuvre, or unpredictable weather conditions, all of which interact with real vehicles.
These digital elements behave just like real-world dangers, allowing the reactions of autonomous systems to be examined in a safe, repeatable and cost-effective manner. The Discovery series shows how testing first takes place in the laboratory using scaled-down versions of real cars, before later moving on to full-size vehicles on a test track.
Watch the episode in late December
In the 28 December episode, Csaba Gyetván also takes on a model car controlled not by a pre-programmed algorithm, but by reinforcement learning. The car – or rather, its algorithm – has effectively trained itself, having spent around 3,000 hours driving in a simulator.
Reinforcement learning works much like training: the algorithm received virtual “rewards” for driving efficiently and safely, and “penalties” when it made mistakes. The question remains: will the human or the machine emerge victorious from the race?