16 Questions About Self Driving Cars | Summary and Q&A

TL;DR

Self-driving cars are on the horizon, and this analysis delves into technology, business, and social implications surrounding their development.

Key Insights

• 😨 The battle for self-driving car dominance may be between incumbents and native Silicon Valley companies, with the Chinese market also becoming significant.
• 😨 The concept of transportation as a service may reduce car ownership, impacting various aspects of the industry, from insurance to car repairs.
• 😨 Accident rates are expected to decrease significantly with the introduction of self-driving cars, with potential short-term challenges in the transition phase.
• 🪛 The shift towards self-driving cars could have profound social implications, from changes in city infrastructure to questions about liability and the role of driving in society.
• ⌛ Commute times could potentially increase due to a shift in emphasis from driving experience to productivity and leisure during the commute.
• 🥺 The introduction of self-driving cars may lead to unforeseen second and third-order effects, similar to the impact Walmart had on the retail industry.
• 😨 Estimates for the widespread availability of self-driving cars vary greatly, with predictions ranging from 2018 to 2040.

Transcript

Read and summarize the transcript of this video on Glasp Reader (beta).

Questions & Answers

Q: Will self-driving cars be introduced incrementally or immediately at higher levels of autonomy?

The approach to introducing self-driving cars can vary, with incumbents likely taking an incremental approach, adding features one at a time, while companies like Google may opt for a higher level of autonomy to avoid user experience design challenges.

Q: Is lidar necessary for self-driving cars, or can stereo cameras provide similar functionality?

Opinions on the necessity of lidar differ, with some arguing that stereo cameras can achieve similar results in terms of 3D mapping, while others highlight the benefits of lidar's accuracy and resolution.

Q: Will self-driving cars rely on new precomputed HD maps, and if so, who will provide them?

The use of new HD maps specifically for self-driving cars is being explored, but the question of who will provide them and the potential limitations they may impose on the mobility of self-driving cars remains uncertain.

Q: What blend of software techniques will be used in self-driving cars?

There is a debate between those advocating for deep learning algorithms and those who argue for a combination of traditional robotic control systems and algorithmic techniques.

Summary

In this video, Frank Chen from Andreessen Horowitz discusses the future of autonomous cars and answers 16 questions related to technology, business, and social implications. He talks about the categorization scheme for self-driving cars, the debate surrounding the use of lidar as a sensor, the need for new HD maps, the role of deep learning in self-driving cars, the use of real-world versus virtual reality data for training, the importance of V2X radios, the possibility of eliminating traffic lights, the localization of self-driving cars to different driving cultures, the changes in the automotive industry and car ownership models, the impact on insurance and liability, the evolution of accident rates, the potential increase in commute times, the unforeseen social consequences of autonomous cars, and the estimated timeline for widespread adoption of autonomous cars.

Questions & Answers

Q: What are the different levels of categorization for self-driving cars?

The Society for Automotive Engineers (SAE) has a six-level categorization scheme, ranging from Level 0 (no automation) to Level 5 (full automation). The United States has its own slightly different categorization, but for this international audience, let's focus on the SAE scheme.

Q: How will we get to the self-driving future?

This question refers to whether we will gradually add features to existing cars or make a sudden leap to fully autonomous vehicles. Incumbent auto manufacturers tend to favor the incremental approach, adding one feature at a time. However, companies like Google may opt to go straight to a higher level of autonomy to avoid user experience challenges.

Q: Will lidar be used as a sensor in self-driving cars?

Lidar, which stands for Light Detection and Ranging, creates a 3D map of the environment using lasers. The use of lidar is a divisive issue in the autonomous vehicle ecosystem. While its accuracy and resolution are beneficial, its cost and intrusive design are concerns. Some argue that stereo cameras can provide similar benefits without the additional cost and complexity of lidar.

Q: Will self-driving cars require new types of precomputed HD maps?

Existing mapping services like Google Maps do not provide enough information for self-driving cars to navigate safely. HD maps, specifically designed for autonomous driving algorithms, could include details such as curves, traffic barrels, and glare information. The question is whether these maps are necessary and who will provide them.

Q: What blend of software techniques will self-driving cars use?

The current hot technology in Silicon Valley is deep learning and neural networks. Deep learning, combined with end-to-end training, is believed by some to be the way to achieve fully autonomous driving. However, others argue that a combination of deep learning and traditional robotic control techniques is necessary for reliable vehicle control.

Q: How much real-world data versus virtual reality data will be used to train self-driving cars?

Deep learning algorithms require a significant amount of data for training. Some propose using simulators and gaming engines to simulate real-world driving scenarios and train the algorithms. The advantage is that simulations can expose the algorithms to a wide variety of situations. The question is whether virtual reality data can fully replace real-world data.

Q: Will V2X radios play an important role in self-driving cars?

V2X radios, which include vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, can enable cars to communicate with each other and with road infrastructure. For example, cars could coordinate at intersections or receive signals from traffic lights. While they offer exciting possibilities, widespread deployment and protocol compatibility present challenges.

Q: When will we be able to eliminate traffic lights?

In a world dominated by autonomous cars, traffic lights may become obsolete. Through communication and coordination, intersections can be managed dynamically, eliminating the need for traditional traffic lights. This could result in smoother traffic flow and efficient use of space.

Q: Will self-driving cars be localized to specific driving cultures or adaptable to various environments?

There is a debate over whether self-driving cars should be programmed to understand and adapt to specific driving conventions in different cities or learn on the go. The idea is to train the algorithms to understand local driving behavior, which can vary greatly. This could be achieved through data collection and machine learning.

Q: How will the way money moves change with the rise of self-driving cars?

One question is which players will dominate the self-driving car market. Will it be the incumbent auto manufacturers, native Silicon Valley companies, or Chinese manufacturers who aggressively pursue this technology? Additionally, there is a discussion about whether consumers will continue to buy cars directly from manufacturers or shift towards transportation as a service, which would change the dynamics of the industry.

Q: Will self-driving cars lead to a shift from car ownership to transportation as a service?

If transportation shifts towards a service-based model, car ownership could become less essential. Consumers might prefer to use transportation services like Lyft and Uber rather than owning their own cars. This shift could have significant implications for the auto industry, insurance, and consumer loyalty.

Q: How will repair costs be affected by self-driving cars?

It is anticipated that self-driving cars will have fewer accidents, but repairs might be more expensive due to the complex technology involved. The cost of repairing the onboard supercomputer responsible for autonomous driving might offset the reduction in accident rates. Additionally, the issue of liability arises, as determining responsibility for accidents involving autonomous cars could be complicated.

Q: How will accident rates change with the introduction of self-driving cars?

While fully autonomous cars are expected to significantly reduce accidents, there might be a temporary increase in accident rates during the transition phase when autonomous and human drivers coexist on the roads. The learning curve for autonomous vehicles might result in new types of accidents as they interact with human-driven cars. However, the long-term goal is to reduce accident rates to almost zero.

Q: How will commute times be affected by self-driving cars?

There are conflicting predictions regarding commute times. On one hand, if people become indifferent to their commute time since they don't have to actively drive, it could lead to longer commutes. On the other hand, the availability of self-driving cars could enable efficient use of city spaces, reducing congestion and shortening commute times.

Q: What are the social consequences and second order effects of self-driving cars?

Just like the invention of cars had unforeseen consequences like the rise of Walmart, self-driving cars will have second and third order effects that are difficult to predict. Changes in car ownership and the transportation landscape will impact various industries and social norms. It remains to be seen how self-driving cars will transform society.

Q: At what point will it become illegal to drive manually?

If self-driving cars are statistically proven to be safer and more efficient than human drivers, there might come a point where manual driving becomes illegal. Once autonomous vehicles demonstrate superior safety records, there would be no need for human drivers on public roads.

Q: When will self-driving cars become a widespread reality?

Different players in the industry have different timelines for the widespread adoption of self-driving cars. Estimates range from companies like NuTonomy predicting availability by 2018 to the IEEE projecting that by 2040, 40% of cars on the road will be autonomous. The timeline will depend on technological advancements, regulatory frameworks, and market demand.

Takeaways

The journey towards a future of autonomous cars is underway, and there are many technical, business, and social questions to address. The adoption of self-driving technology will depend on a mix of incremental improvements and potentially disruptive approaches. The use of lidar and advanced mapping systems, the blend of software techniques, the role of V2X radios, and the integration of self-driving cars into the urban infrastructure are all significant aspects to consider. Additionally, changes in car ownership models, insurance, liability, accident rates, commute times, and broader societal impacts will accompany the widespread adoption of self-driving cars. The timeline for this future is still uncertain, but it is clear that autonomous cars will have a profound impact on our lives in the coming years.

Summary & Key Takeaways

• The future of self-driving cars can be achieved either incrementally, adding new features over time, or by immediately introducing level 4 or 5 autonomous capabilities.

• The debate surrounding the use of lidar sensors in self-driving cars continues, with arguments for both its necessity and the possibility of achieving similar functionality with just stereo cameras.

• The creation of new types of precomputed HD maps specifically for self-driving cars is being explored, but concerns remain over who will provide these maps and the potential limitations if they are not available.

• There is a debate over the use of deep learning algorithms versus traditional robotic control systems in self-driving cars, with potential benefits and challenges associated with both approaches.