Key Takeaways
1. XPeng has developed its own AI driving chip called Turing, enabling a vision-only self-driving system.
2. The company plans to launch a Level 4 autonomous robotaxi next year, easier to develop than their current Level 2+ driver-assist system.
3. The Level 4 system can make quick decisions without driver intervention, unlike Level 2 systems that must consider unpredictable human behavior.
4. Driver emotions significantly influence the need for technology to align with human thought processes, especially in critical driving situations.
5. XPeng’s self-driving technology relies on machine learning and cameras, eliminating the need for LiDAR, while they await regulatory approval for deployment in regions like Europe.
One of the leaders in self-driving technology among Chinese electric vehicle manufacturers, XPeng, has developed its own AI driving chip called Turing. This innovation enables the company to implement a vision-only system, leveraging recent improvements in machine learning. Their driver-assist feature, named XNGP, stands for XPeng Next Generation Pilot, and it is specifically designed for self-driving in dense urban environments.
Future Plans for Robotaxi
As per Candice Yuan, head of XPeng’s self-driving division, the firm plans to introduce a dedicated robotaxi akin to Tesla’s Cybercab next year. This vehicle will utilize Level 4 autonomy, which is similar to Tesla’s Full Self-Driving (FSD) without supervision. Interestingly, XPeng found that developing the Level 4 system, which doesn’t require any driver intervention, was simpler than the current Level 2+ XNGP system that their vehicles are operating on.
Decision-Making Capabilities
Yuan emphasized that the Level 4 system can make rapid decisions, paralleling the unsupervised FSD employed by Tesla’s pilot robotaxi service in Austin. Conversely, the Level 2 autonomy has to consider the unpredictable nature of human behavior, which can often interfere with the self-driving software’s logical operations. She provided several scenarios where the AI system might assess a situation as safe, but a human driver might feel uneasy and decide to take control:
For instance, drivers prioritize efficiency. If the Advanced Driver Assistance Systems (ADAS) perform correctly but the vehicle is slower than anticipated, a driver may intervene. In another scenario, the system might deem a situation safe to proceed, but the driver could perceive it as too tight or risky and take over. Lastly, if the vehicle accelerates either too quickly or too slowly, the driver might feel uncomfortable and regain control. Consequently, achieving a higher level of autonomous driving (Level 2) for passenger cars appears more complex than the Level 4 system that Alibaba developed for driverless vehicles.
Emotions in Driving
Drivers are influenced by their emotions and feelings, so it’s crucial for the technology to align with their thought processes. The experience matters greatly. For example, if a robotaxi travels off course, it will continue along that path until it can turn around and correct itself. However, in Level 2 scenarios, a driver is likely to take over if the vehicle strays from the intended route; therefore, going the wrong way is not acceptable.
XPeng’s Level 4 autonomous driving system is built to learn from short video clips showing both effective and ineffective driving methods, similar to Tesla’s unsupervised FSD. The company indicated that advancements in machine learning allowed them to eliminate LiDAR, relying solely on cameras installed in their vehicles for self-driving. However, third-party chip manufacturers were unable to provide the necessary customization for XPeng’s XNGP system to integrate its vehicle hardware with the in-house developed ADAS software, motivating the automaker to create its own specialized AI chip.
The large language model employed by XPeng’s vision-based system can be utilized in various applications, akin to Tesla’s FSD. XPeng is now just waiting for public regulatory frameworks to catch up before deploying its self-driving technology in regions like Europe, where they recently initiated localized electric vehicle manufacturing to bypass import tariffs.
Source:
Link
Leave a Reply