Bosch Global Software Technologies Private Limited is a 100% owned subsidiary of Robert Bosch GmbH. Bosch Global Software Technologies is one of the world’s leading global suppliers of technology and services, offering end-to-end Engineering, IT, and Business Solutions. Bosch Global Software Technologies is at the forefront of designing, developing, and executing IoT ecosystems through our all-encompassing capability within the 3 aspects of IoT – Sensors, Software, and Services.
In a recent interview, Abdullah interacted with Bindu Santha Philip, Vice President Technology , Bosch Global Software Technologies in which he discussed technical development of autonomous driving, obstacles preventing autonomous driving technologies from being widely adopted, systems for autonomous driving adjust to different circumstances and environmental conditions, machine learning and artificial intelligence be used to improve the capabilities of autonomous driving, moral and ethical issues figure into decision-making scenarios while using autonomous driving systems.
Which major technical developments are propelling the development of autonomous driving?
Artificial intelligence (AI) is revolutionizing the world, with data-driven AI leading the charge towards autonomous driving. However, achieving the full potential of AI requires the convergence of several critical and enabling technologies. The key technical developments propelling this transformation include:
- New age E/E architectures to converge technology landscape
The vehicle E/E architecture will transition from distributed (smart sensors) to centralized architecture (vehicle computer), as the distributed system cannot meet the demand of data driven AI computation capability, communication efficiency and greater bandwidth capacity. - Separation of hardware and software
Hardware and software abstraction are critical tools for managing the complexity of autonomous driving systems. The autonomous software needs to be modular, loosely coupled, and provide abstraction for hardware and operating systems. - Synergies among Compute, Data, Analytics & AI
Accurate environment perception and behaviour estimation of vehicle ecosystem participants are crucial building blocks for autonomous driving systems. Advancements in sensor technology, computing capabilities, and data-driven AI development are essential to realize these components. However, the maturity of these building blocks is proving to be more challenging than anticipated, as evidenced by delays in the release of autonomous vehicles. As these foundational technologies (sensing, processing, and data management) continue to evolve, computing requirements are expected to be at the forefront of automotive engineering. Consequently, powerful processors, custom-designed AI chips, power management, and software/data optimization will be critical in making autonomous vehicles a reality. - Software updates
An incremental approach (ex. AI models which can handle 90% of the scenarios, then improving the performance to handle the rest of the 10%) offers a more cautious and commercially viable path towards AD. To attain autonomous capability, Over-the-air (OTA) updates will be integral part of autonomous vehicles. OTA technology will improve safety functionalities, better maneuvering in complex situations (adapt to new traffic rules, traffic regulations, construction zones, diversions) and robust cybersecurity measures to prevent unauthorized access. Overall, OTA is a game-changer, it ensures continuous improvement, enhances safety/security, and paves the way for an evolving AV.
What are the primary obstacles preventing autonomous driving technologies from being widely adopted?
Demographic, economic, and traffic factors all influence the adoption of autonomous driving technologies. Here are some of the key hurdles preventing these technologies from being widely adopted:
- Sensor Limitations:
AV sensor technology have limitations in adverse weather conditions, unfamiliar objects, edge case handling as well as complex/chaotic maneuvers. Reliable perception in all environmental condition is crucial for AV. - Cost of Technology:
The development, manufacturing, and implementation of autonomous vehicle (AV) technology are costly. Additionally, a large-scale transformation in human resources and their skills is required for the industrialization of AV technologies. Beyond the high initial costs, concerns about safety and the overall value proposition also limit consumer adoption rates. - Social acceptance:
Most of AV technologies are built around AI. Social acceptance of AI technology is a concern. The societal fears, lack of transparency, explainability, bias/ fairness and ethical issues might decelerate adoption of AI technologies. - Robust Regulations Frameworks:
A robust regulatory and ethical framework is essential for building trust in autonomous vehicle (AV) technology. These frameworks, acceptable to all stakeholders, are necessary to address liability in case of accidents and define responsibilities for cybersecurity, safety, and ethical issues. Additionally, current regulations are not fully adapted for the use of AVs on public roads.
Overall, these obstacles are complex and interconnected. Overcoming them will enhance wide adaptation of autonomous driving technologies.
How do systems for autonomous driving adjust to different circumstances and environmental conditions?
The key task of an autonomous vehicle (AV) is to navigate safely and efficiently under all conditions, including day and night, and various weather and traffic scenarios. This is achieved through the principle of sense, think, and act.
- Sense: The most crucial task for an autonomous vehicle (AV) is to gather real-time information about its static and dynamic surroundings. To achieve this, data from various sensors (such as radar, video, NRSC, USS, lidar, and maps) are fused. This fused data helps the AV detect, track, and classify objects (e.g., vehicles, pedestrians, roads, lanes) and understand their positions and movements.
- Think: Based on the perceived environment, AV software predicts the intentions of detected objects and makes real-time decisions to navigate safely. The “think” functions must determine the route, trajectory, and safe maneuvers based on real-time data and perceived or predicted objects and conditions.
- Act: A closed-loop control system for various actuators (steering, acceleration, braking, etc.) enables precise movement of the autonomous vehicle. To manage complexity in different circumstances and environmental conditions, AVs rely on data-driven development. Data from various sources (sensors, traffic participants, simulators, infrastructure) is used to train and improve the vehicle’s perception (sense), dynamic replanning (think), and overall performance.
How can machine learning and artificial intelligence be used to improve the capabilities of autonomous driving?
Data is the most important asset for AV’s. Data extraction and analytics plays a vital role to bridge the performance gap between human driver and AV’s.
The big data (Volume, Variety and Velocity) of AV, requires AI technologies.AI can improve data processing efficiency by automating various data science tasks (data labelling, cleaning). The processed data is fused by algorithms to extract a unified understanding of vehicle perception (object detection, tacking, classification, and prediction). AI techniques like deep learning can handle complex data types (point cloud, images) to build or enhance perception of AV’s.
Going forward, the perceived vehicle and environment data can be used by AI/ML algorithms to make safe driving decisions. (Determine route, trajectory and safe maneuvers). AI model can learn from experience of other vehicle or environment, to improve their performance and achieve safe navigation on the roads.
In order to reduce vehicle validation effort, AI can be used to create a virtual driving scenario, including edge cases that might be difficult to replicate in the real world. With the help of AI, edge case or validation scenarios can be simulated on digital twin of the physical world. Also, Usage of Gen AI can further improve software development efficiency.
In short AI/ML is critical technology for autonomous driving.
How do moral and ethical issues figure into decision-making scenarios while using autonomous driving systems?
Traditionally, decision-making scenarios of autonomous driving systems will be built using AI/ML models. The decision-making process of these AI/ML models is opaque, and it can be difficult to understand input / output correlation. A biased data or algorithmic could bring social inequalities, unfairness and inaccuracy. In addition, the maturity of AI technology may also impact the decision making where technology shall distinguish absolute or contextual information. For example in a search algorithm a query “find image frames of large humans” could result in images frames with fat human beings, however, intent of the query is to find images in which humans are closer to camera. This kind implicit contextual information may introduce an unintentional bias in data. A discriminatory data, algorithm and lack of transparency in decision-making becomes crucial for moral, ethical, and social acceptance of AI technology. Going forward, indicators representing moral and ethical impact of AI technology may influence regulatory requirements or social acceptance of the AI technology.