The Evolution of AI in Mobility & Robotics
The way machines perceive the world around them has been transformed by the rise of artificial intelligence. Over the past decade, breakthroughs in deep learning and available computational power have unlocked progress in computer vision and robotics.
Research in autonomous driving is continuously moving us closer to safer roads, smarter vehicles, and more intuitive interaction with machines. From lightweight neural networks that can detect vehicles on resource-constrained platforms, to advanced models that combine camera images with depth information for predicting the movements of cars and pedestrians, AI is becoming the backbone of modern mobility.
Advanced computer vision solutions also open the door for humanoid robots to step into real-world environments as social agents guiding pedestrians, signaling to drivers, and acting as dynamic mediators in traffic.
Advancing Perception Through Synthetic Data and Simulation
One of the most frequently implemented AI solutions in Advanced Driver Assistance Systems (ADAS) are image classification and object detection, including traffic signs. Vast majority of AI-based systems is highly dependent on high quality and robust training datasets. Since collecting suitable data in all environmental conditions requires a lot of effort, researchers often utilize synthetic modifications of existing data. Our paper ‘**Evaluating a Siamese Network for Traffic Sign Recognition using Synthetic Datasets’** proposes a set of hand-crafted image augmentation methods to create synthetic traffic sign datasets based on Slovak road sign templates. The resulting synthetic datasets were then used to evaluate a Siamese neural network, a model designed to distinguish between similar and dissimilar inputs. When even synthetic modifications of real-world data fail to provide your AI-based system with enough data to be sufficiently robust, data generation and simulation can save the day. We address this as well in our research **’ROSBENCH: A Simulation-Based Benchmark for Sensor Quality and Environmental Conditions Robustness in AV Perception’** which demonstrates how you can make your solutions ready for a vast array of different conditions without the headaches of extensive real-world data collection.
Sensor Fusion and Lightweight Models for Real-Time Understanding
Perceiving the road and participants on the road is essential, but in order to ensure safety on the roads, systems have to understand the traffic situation and even predict the next steps. In order to do so, data from RGB cameras is often enriched with depth information from sensors like LiDAR or radar - performing sensor fusion. The **Li-ViP3D framework** integrates camera and LiDAR data to enhance end-to-end perception and prediction. Instead of treating detection, tracking, and forecasting as separate tasks, Li-ViP3D unifies them in a single differentiable pipeline. The end result is a substantial improvement in predicting the trajectories of cars, cyclists, and pedestrians and better accuracy for long-horizon forecasts.
Advanced data fusion opens new horizons and delivers impressive results, but at the cost of significant computational burden. To enable real-time processing on limited hardware, we explored more lightweight approaches. Specifically, in the paper ‘**Single-filter CNN for Vehicle Recognition’**, we proposed a simple method capable of detecting the presence of an object of interest (namely, a vehicle) using only a single convolutional kernel. In another experiment, we investigated the impact of basic post-processing techniques, such as bounding box rotation based on LiDAR point proximity, on improving the accuracy of lightweight, CPU-runnable 3D object detection models.
Robotics in Real-World Traffic Scenarios
Computer vision solutions for mobility and robotics are no longer just limited to controlled experiments in perfect conditions. We have been exploring their practical applications in real-world conditions. Specifically, we’ve collaborated on research focused on how humanoid robots can directly intervene in traffic situations to assist vulnerable road users. We’ve enhanced the perception and decision-making capabilities of the ARI, social robot platform, to serve as a crossing assistant. Equipped with upgraded sensors and object detection algorithms, ARI can evaluate vehicle behavior, recognize pedestrian intent, and signal when it is safe to cross. This proof-of-concept demonstrated how social robots can act as trusted mediators between humans and vehicles, improving safety and efficiency in complex urban traffic environments.
We took this vision further in another study by evaluating ARI at two-lane pedestrian crossings, where it manages interactions step by step. Running real-time perception models on NVIDIA Jetson AGX Orin, the robot detects pedestrians and approaching vehicles, then guides people across with clear gestures and voice prompts. By leveraging depth information and object tracking, ARI ensures that pedestrians cross the first lane safely before guiding them across the second. This structured process replicates the role of human traffic police, while providing continuous monitoring and support.
At NetFire, we see these breakthroughs in autonomous perception, robotics, and sensor fusion as part of a larger ecosystem of intelligent systems. Our mission is to help organizations turn cutting-edge research into practical, scalable, and secure AI deployments. From simulation-based dataset generation to multi-sensor perception pipelines and robotic mediation frameworks, we provide the infrastructure and expertise to bridge the gap between proof-of-concept and real-world deployment.
How to learn more or get in touch
- Visit our Resources page to get the latest NetFire product news, company events, research papers, branding guidelines, and much more.
- Explore our Support Center for overviews and guides on how to use NetFire products and services.
- For partnerships, co-marketing, or general media inquiries, email marketing@netfire.com.
- For all sales inquiries, email sales@netfire.com to get setup with an account manager.
