New Delhi: Artificial Intelligence (AI) is revolutionising the automotive industry, ushering in an era of enhanced safety, efficiency, and user experience.
The automotive industry is increasingly adopting AI technology to streamline operations and improve the overall vehicle performance. By tapping into the potential of big data, IoT, AI, and ML, artificial intelligence has completely transformed how vehicles are designed, manufactured, and driven. From autonomous vehicles to advanced safety systems, the advantages of AI in the automotive industry are enormous.
AI in ADAS and Autonomous Mobility
Autonomous Driving has transcended far beyond being a moonshot idea over the last half-decade or so. Over the years, its popularity has increased as it promises to shape the future of mobility.
AI is one of the main reasons for the rapid progress of Autonomous driving (AD) as it relies heavily on AI to perceive the environment, make decisions, and control the vehicle.
AI Algorithms in ADAS: Sensing, Planning, and Action
Sensing & Perception:
Perception Algorithms: AI-powered perception systems use sensors like cameras, LIDAR, radar, and ultrasonic sensors to interpret the vehicle’s surroundings. Convolutional Neural Networks (CNNs) and Deep learning algorithms are widely used for object detection, classification, and tracking, enabling the vehicle to recognise other vehicles, pedestrians, traffic signs, and obstacles.
Cameras, Radars, Lidars along with Ultrasonics are the primary sensors that are used for autonomous mobility as it demands high performing sensors. For example, Cameras are increasing their capabilities in terms of their resolutions, extended FOV’s (field of view), low light capabilities, stereo, HDR etc. Roads are designed for human eyes and hence cameras are important sensors that detect the road semantics.
Earlier classical CV methods were used for sensing or perception functions. Continental is a leading supplier of high performing, modular and robust camera sensors that are scalable with option of having smart camera to support NCAP (New Car Assessment Program) ADAS functions and Satellite cameras to scale up from *L2+ to L4/L5 AD systems. Additionally, radars, lidars and ultrasonics are increasing the capabilities.
For example, Multi channel along with elevation also called 4D sensing Radars, Chirp chirp transformations, micro doppler etc. are being introduced. For LIDAR, MEMS (Micro Electro Mechanical Systems) based LIDAR, configurable wide and long range Lidars and ultrasonics with enhanced detection capabilities from near range to farther ranges have been introduced. Continental is also a leading supplier of high performance 4D premium long range radar sensors and short range radars which enables highly automated driving along with EBA (Emergency Braking Assist), ACC (Adaptive Cruise Control), Lane change assists, and Blind spot warnings.
All these sensors contribute to rich data that is useful for AI/Deep learning applications to further augment perception performance which hitherto was hitting a limit due to classical approaches.
Sensor Fusion: Sensor fusion is critical in applications such as autonomous driving, robotics, and surveillance, where multiple sensors are used to accurately perceive the environment. By combining data from multiple sensors, AI algorithms can create a comprehensive understanding of the environment. In sensor fusion, most advanced AI algorithms like BEV (Bird Eye View) are being used. The BEV algorithm transforms sensor data into a top-down view of the surroundings, which can then be used for various perception tasks such as object detection, lane detection, and obstacle avoidance.
Planning:
Path Planning Algorithms: AI algorithms help calculate the safest and most efficient route for the vehicle. These algorithms, enable real-time decision-making to navigate complex environment.
Action:
Control Algorithms: AI-based control systems execute the planned path by managing the vehicle’s steering, acceleration, and braking. Control algorithms, such as Model Predictive Control (MPC) and Proportional-Integral-Derivative (PID) controllers, are optimised using AI techniques to maintain vehicle stability and passenger comfort.
AI, SDV & OTA:
With the advent of Software-Defined Vehicles (SDVs) and Over-the-Air (OTA) updates, AI models are continuously being refined and improved based on real-world data and fleet performance. This dynamic process involves collecting data from vehicles in operation, where any issues or edge cases encountered are recorded and analysed. Engineers then use this data to identify patterns, fix bugs, and address previously unforeseen scenarios. The AI model is subsequently retrained using this new data, incorporating the latest learnings to enhance its accuracy and robustness. Once updated, the refined model is deployed back to the fleet via OTA updates, ensuring all vehicles benefit from the improvements without the need for manual interventions. This iterative cycle not only enhances the safety, efficiency, and reliability of SDVs but also ensures that the vehicles remain at the forefront of technological advancements, adapting swiftly to new challenges and environments.
AI in Driver Monitoring Systems & Deviceless Access
At Continental, the interior of the vehicle is the primary focus, offering the customers more than just safety and comfort. Driver monitoring systems (DMS) use AI to enhance vehicle safety by assessing the driver’s state. AI algorithms analyse data from in-cabin cameras and sensors to detect signs of drowsiness, distraction, or impairment. Techniques such as facial recognition, eye-tracking, and behavioral analysis enable real-time monitoring and alert systems to warn the driver or take corrective actions if necessary.
Keyless entry is another application of AI. Authentication is done through the existing sensors present on the vehicle. Different authentication methods include iris, voice, fingerprint and face recognition. With the help of AI, issues like multi enrolment, gait patterns, remote delegation, remote unlock etc. are being resolved.
AI Transforming the User Experience:
Predictive Maintenance:
AI algorithms analyse vehicle data to predict component failures and maintenance needs. Machine learning models can identify patterns in sensor data to forecast issues before they occur, reducing downtime and maintenance costs.
Battery Management Systems (BMS):
For electric vehicles (EVs), AI-enhanced BMS optimise battery performance and longevity. AI models predict battery life, manage charging cycles, and balance cell loads to ensure efficient and safe battery usage.
Infotainment Systems:
The integration of AI has made the infotainment systems smarter. Large Language Models (LLMs) like GPT-4 (Generative Pre-Trained) are used to enhance voice assistants, enabling more natural and intuitive interactions with the users. AI-powered infotainment systems can provide personalised recommendations, real-time navigation assistance, and seamless connectivity with other smart devices.
Voice Assistance:
While automobile companies often prefer implementing third-party voice assistants, some industry players choose to build their own voice-recognition software. Such AI-enabled personal assistance in cars helps make calls, adjust the temperature, change radio stations, play music, inform about the gas amount in the tank, and do a lot more.
Most importantly, voice recognition tools have high personalisation capabilities, meaning they can remember the users’ interests and advise adjustments based on their history.
Passenger Experience:
AI aims to transform the overall driving experience by providing safety and comfort to the driver as well the passengers.
Considering the passengers’ experience and safety on the road, automotive manufacturers strive to upgrade their vehicles with technologies like IoT, image data, NLP, and object identification.
It allows the passengers to make specific commands and listen to their favorite music, order food while enjoying their journey on the road.
Cost Savings:
The adoption of AI in the automotive industry significantly helps reduce costs in all aspects of operations, from designing to manufacturing.
By optimising manufacturing processes, improving supply chains, and identifying potential issues in vehicles, AI can help reduce costs in various ways.
Challenges in Implementing AI in Automotive
Safety Challenges:
Reliability: Ensuring the reliability of AI systems in diverse driving conditions is a critical challenge. AI models must be trained on vast datasets covering various scenarios to perform accurately in real-world situations. There are always edge cases that the AI model would not have been trained or have encountered or predicted.
Security Challenges: Adversarial Attacks
Adversarial attacks can challenge the resilience of autonomous driving systems (ADS) by misleading their deep neural networks (DNNs) into incorrectly classifying traffic signs. These attacks can be malicious or natural and can be performed using printed signs or stickers in the real world. Adversarial attacks are conducted by adding tiny but malicious perturbations, these perturbations can fool DNNs with high probability.
Attacks can be digital attacks, visible attacks, invisible attacks and physical attacks.
For an AI system, one needs to take care of Data issues and Model issues. Data issues involve issues like labelling consistency, data outliers, data sufficiency, data leakages, dataset balances. Model issues can be handled with rigorous test frameworks like Failure model analysis, Confusion matrix tests, Hard positive/Negative tests, white noise tests and adversarial robustness tests, bias and fairness tests.
Explainability:
Traditional algorithms like decision trees, Bayesian nets are explainable but are less accurate. Deep Learning networks are highly accurate but are less explainable. Deep learning networks used in self-driving cars are so deep that it is very difficult to understand what is going on in each layer, making it hard to pinpoint the fault. Thus, explainability is important from psychological, socio-technical, legal, and philosophical perspectives.
In practice, autonomous driving systems often use a blend of methods (interpretable models, such as decision trees or rule-based systems, saliency maps, LIME – Local Interpretable Model-agnostic Explanations) to provide transparency about their actions, such as explaining why a car took a particular route, slowed down, or stopped. This transparency helps engineers troubleshoot and improve the systems, ensures that the decisions made align with safety standards, and builds trust with users and stakeholders.
To drive desirable outcomes with explainable AI, one needs to consider the following:
Fairness and debiasing: Manage and monitor fairness. Scan the deployment for potential biases.
Model drift mitigation: Analyse the model and make recommendations based on the most logical outcome. Alert when models deviate from the intended outcomes.
Model risk management: Quantify and mitigate model risk. Get alerted when a model performs inadequately. Understand what happened when deviations persist.
Lifecycle automation: Build, run and manage models as part of integrated data and AI services. Unify the tools and processes on a platform to monitor models and share outcomes. Explain the dependencies of machine learning models.
Multicloud-ready: Deploy AI projects across hybrid clouds including public clouds, private clouds and on premises. Promote trust and confidence with explainable AI.
Additionally, Continental is collaborating with a startup with their comprehensive AI testing platform that automatically detects issues, identifies root cause leading to a 3x acceleration in AI development lifecycle while reducing AI risk exposure by 90% in production.
The integration of artificial intelligence into the automotive industry represents a transformative leap forward, driving advancements that were once considered to be fictional. AI’s role in enhancing vehicle safety, efficiency, and user experience is undeniable, with autonomous mobility standing at the forefront of this revolution. From predictive maintenance and intelligent traffic management to personalised in-car experiences, AI is reshaping the way one conceives and interacts with vehicles.
Looking ahead, the connection between AI and automotive innovations promises a future of unprecedented mobility solutions. By continuing to push the boundaries of what is possible, the industry is paving the way for a safer, more efficient, and interconnected world. The journey towards fully autonomous, AI-driven vehicles is a testament to human ingenuity and the relentless pursuit of progress. Thus, the potential for positive change is vast, heralding a future where transportation is smarter, more sustainable, and more human-centric.
(Disclaimer: Bhanu Prakash is Head of Product Lines: Camera and ADCU, Autonomous Mobility, Continental Automotive India. Views are personal.)
