Pony.ai’s Post

Really happy with the outcome from this GSoC project. You can learn more about the project in her blog: https://lnkd.in/dw3vYfwW Video: https://lnkd.in/d8KZhEx5

Board’s Technology Oversight Alert!! I encourage lead Director and Chair of Nomination and Governance Committee to refresh their Board with deep technology expertise as it has direct impact to value creation “and” business resilience. Take a look at Missy Cummings‘s insights here.  ✅ The lack of the technical comprehension across industry and government is appalling. A language model may give you nonsense, a self-driving car can kill you.  ✅ Human errors in operation get replaced by human errors in coding. ✅ AI failure modes are hard to predict. ✅ Probabilistic estimates do not approximate judgement under certainty. ✅ Maintaining AI is just as important as creating AI. ✅ AI has system-level implications that can’t be ignored. ✅ Neither AI in LLMs nor the one in autonomous cars can understand the situation, the context, or any observed factors. Christina Montgomery Ruchir Puri Kyle Brown James Lam Francesca Rossi Gary Marcus IEEE #CEO #KSgems #CIO #CTO #CISO #AI #AIsafety #IEEESpectrum

The Countdown - Top 100 Companies (in no particular order). # 72 Leveraging AI for Tomorrow's Challenges: Insights from Leading UK Start-ups Company: "FiveAI: Leading Autonomous Vehicle Technology" Website: https://www.five.ai/ Video: https://lnkd.in/eCYSParJ LinkedIn: https://lnkd.in/eJtNyd2N What FiveAI Does: FiveAI, based in the UK, is a trailblazer in developing autonomous vehicle technology. They focus on creating advanced software for self-driving vehicles, aiming to build Europe's automated driving champion. Their work involves tackling complex problems in computer vision, machine learning, and robotics to deliver autonomous transport solutions. AI Initiative: FiveAI has formed a partnership with Bosch, emphasizing their commitment to advancing autonomous vehicle technology. Additionally, they have collaborated with Cognata to develop a comprehensive verification approach for autonomous vehicle systems. This approach minimizes the need for costly real-world testing and accelerates the development of automated driving systems (ADS) features. Impact: FiveAI's initiatives in autonomous vehicle technology are poised to significantly impact the future of urban mobility. Their partnerships and innovative verification approaches are key steps in bringing safer, more efficient autonomous driving solutions to the market. These advancements are critical in shaping the future of transportation and urban planning. Trends in this field: The autonomous vehicle industry, as exemplified by FiveAI, is rapidly evolving with a focus on integrating AI and machine learning into transportation solutions. There is an increasing trend towards developing advanced verification systems for autonomous vehicles, reducing reliance on real-world testing and speeding up innovation. Collaborations between tech companies and automotive leaders are becoming more prominent, highlighting the interdisciplinary approach required in this field. Tags: #ai #artificialintelligence #autonomousvehicles #fiveai #transportation #machinelearning #robotics #urbanmobility #automotivetechnology

Cutting-Edge Research Unveils GPT-4V(ision) Breakthroughs in Autonomous Driving #AI #artificialintelligence #Automotive #Autonomousdriving #causalreasoning #Challenges #directiondiscernment #GPT4V(ision) #llm #machinelearning #researchanddevelopment #reshapinglandscape #sceneunderstanding #trafficlightrecognition

🚗 Exciting news in the world of autonomous driving! A comprehensive survey has been conducted on over 200 datasets, providing valuable insights into sensor modalities, data size, annotation quality, and future trends. This study is a crucial guide for developing reliable autonomous driving algorithms and establishing new datasets. #AutonomousDriving #AI #MachineLearning #DataScience #ComputerVision #FutureTech

✅Unveiling the Future of Autonomous Machines. A deep dive into pivotal technologies and platforms propelling the autonomy revolution. Discover insights : https://lnkd.in/d2hCs8_y

We're thrilled to announce significant enhancements to our open-source autonomous driving model comparison tool. We've introduced a brand-new route navigation task. In this challenge, agents must follow high-level commands to navigate a predefined route while avoiding dynamic obstacles. In addition, we've introduced a set of new evaluation metrics. These metrics provide a comprehensive view of the agent's performance, covering everything from route completion ratios to success rates and infractions. Our goal with these enhancements is to provide a robust platform for researchers and developers to benchmark and compare their autonomous driving models. Whether you're working on reinforcement learning, imitation learning, or any other approach, our tool is here to help you fine-tune and evaluate your algorithms. Try out the new route navigation task, experiment with different models, and let us know what you think. Your feedback is invaluable as we continue to evolve this platform. This project was led by Meiqi Zhao through Google Summer of Code 2023 (#GSoC) mentored by Sergio Paniego Blanco, Nikhil Paliwal, and Jose María Cañas from JdeRobot. #AutonomousDriving #MachineLearning #ImitationLearning #OpenSource #GoogleSummerOfCode #GSoC Behavior Metrics: https://lnkd.in/grv3eyAd

#China's 4-stage approach to #EndtoEnd #autonomousdriving (#deeplearning #ai in #EV)

The video explores various AI innovations that are expected to transform people's lives by 2027, including AI-powered medical diagnosis, autonomous vehicles, and personalized education. #technlogy #technews #tech #ai #tools

Human #reasoning seems to be the solution to some of the most popular #ai -use cases theses days, but not the way you expected it.... In a fascinating paper from University of Southern California, Stanford University and NVIDIA, the authors researched if and how human reasoning capabilities could raise the state-of-the-art of current autonomous driving methods which "only" use sensor information and use that to "only"predict and plan where and how to drive, solely based on these sensor inputs. The researchers integrated, and that is the really interesting part, an #llm into the workflow, to serve, so to speak "as the common sense" basis, which is connected to both a task-specific (driving, of course) common sense #memory and a #reasoning engine based on #ChainOfThought reasoning (among other methods). The results are quite interesting, as they in fact show that this system does outperform Sota autonomous driving methods by large margins, some by over 30% 💪 What does this mean? At a minimum: as you might have intuitively guessed, (a) to perform at human-level quality of to even be able to get close to that threshold, autonomous systems such as autonomous cars should incorporate some form of common sense knowledge and reasoning and (b) that LLMs can in fact serve as a basis to mimic that common/world knowledge required for so many human-level quality tasks. What do you think, is this a good trajectory of automation, or is it bad, assuming we will see many more such systems that use LLM-encoded human common sense to automate reasoning/planning and acting in tasks where we would have never expected a machine to get close to our level? https://lnkd.in/dyYfRqjw