The goal of this project is to develop a automated driving system (ADS) that can navigate roads without lane markets in a safe, verifiable manner. As a milestone, this projects aims to develop an Advanced Driver Assistance System (ADAS) following the same principles.
Practically everyone developing autonomous vehicles says autonomous vehicles will be safer, however, there have been some notable accidents involving autonomous vehicles that highlight just how far we have yet to go.1 While these accidents are terrible, by some metrics autonomous vehicles perform better than the average driver. Some experts argue that so long as autonomous vehicles are safer than the average driver, it will result in less injuries and deaths overall and as should be rolled out quickly to not let perfect be the enemy of good.2
As of early 2026, there has been improvements towards a an industry-wide accepted standard for certification; notably the UN's proposed regulation on Automated Driving Systems (ADS)3. These regulations, however, do not define an acceptable level of safety. The most comprehensive currently active regulation is (EU) 2022/14264 which requires system builders to define their own safety critieria but provides an example maximum total system failure rate of 1 fatality per 10,000,000 operating hours.
The current active regulations, particularly (EU) 2022/1426 and (UN) R157, are limited to an Operational Design Domain (ODD) to lined roads in pre-defined areas or paths usually in non-inclement weather. The proposed UN regulation (and many of the standards that influenced it 5) creates a framework for development that requires the creator to define the ODD and what happens when the boundary is crossed. Therefore it is possible to create an ODD for unmarked roads or roads with occluded markings. This system would be developed using the current automotive safety standards such as ISO 26262 and 21448.
Driver-assistance systems may improve driver safety, but are not safety devices. They are typically considered convenience devices as this allows most if not all legal risk to be placed on the operator rather than the manufacturer. Some manufacturers do not make this distinction clear and many operators do not understand these distinctions, sometimes leading to deadly results.1 While operator education may reduce system misuse, it will certainly not eliminate it and the best solution would be to create safe assistance systems.
As we transition to autonomously driven vehicles operated by the likes of Waymo, Zoox, and Tesla, the need for regulation increases. As previously mentioned, these regulations do not state explicit criteria to target, but state that the system creators would need to create their own targets. These targets will likey be determined by the monetary cost of failure and the perceived risk of the system by the general public as manufacturers will be shouldering the legal blame as there is no human operator. Recent research suggests that autonomous vehicles must be 4 to 5 times safer than the average driver to be considered safe by the general public.6
Contemporary AI systems used for self driving cannot be assessed in the same manner as traditional safety systems which can be demonstrated to have a defined output for each input. This is due to the impossibly large range of inputs for vision systems. Instead a series of tests (such as ISO 34501 through 34505) need to be run to demonstrate the efficacy of the system and the failure rates of the system need to be accurately tracked. These are requirements of the testing licenses given to self-driving vehicle companies who wish to drive on public roads.
This system is still in early development and there is no definitive timeline yet. Due to the difficulty of acquiring sufficient ground truth data, development will primarially be done in a simulator. Most simulators are designed for development of highway and urban systems so utilizing one for this system's intended niche is tricky. The current plan is to develop maps and modifications to the CARLA Simulator to facillitate this use case.
Most of the work has gone into understanding the current and proposed regulations and how the ODD can be defined such that bounds of operation can be detected. This includes a potential safety model and how to not accidentially trigger the safety system during what would be considered normal operation.
Furthermore, mapping every unlined road in high definition like what is currently done for self-driving operation is likely an untenable task. Therefore, this project aims to only use navigational map data and sensor inputs to navigate. This has yet to be determined if this information alone is sufficient.