Core Viewpoint - The article highlights the success of the Tsinghua Fire God team in the World Humanoid Robot Games, showcasing advancements in robotics through competitive events, particularly in a 5v5 soccer match where robots operated autonomously [1][21]. Group 1: Event Highlights - The Tsinghua Fire God team won against a humanoid robot version of the German team with a score of 1-0, attributed to a unique "shooting" algorithm that only they had mastered among 50 participating teams [2][25]. - The event featured a variety of competitions, including a 100-meter obstacle race where a robot named Yushu won with a time of 33.71 seconds [8][5]. - The competition included 26 events with a total of 487 matches, showcasing the excitement and technological advancements in robotics [32][33]. Group 2: Technical Insights - The 5v5 soccer match was the first of its kind, with all robots acting autonomously, which increased the complexity of the competition [21][24]. - Each robot was equipped with four cameras for visual perception and distance analysis, allowing them to make quick decisions within 0.1 milliseconds [27][28]. - Strategies varied between teams, with the German team employing a solid defense while the Fire God team utilized a more flexible approach, highlighting the importance of teamwork and algorithmic intelligence in determining the outcome [30][29].

Core Viewpoint - The recent advancements in AI, particularly the launch of the GLM-4.5V visual reasoning model by Zhipu, have garnered significant attention due to its impressive performance in visual benchmark tests, achieving first place in 41 out of 42 tests [2][3]. Group 1: Company Developments - Zhipu has introduced the GLM-4.5V visual reasoning model, which is an open-source model that has excelled in visual benchmark tests [2]. - The GLM-4.5 model has shown substantial improvements in logical reasoning, code writing, and tool invocation [1]. - Despite rapid developments in the AI sector, Zhipu, being a more technology-focused company, has not received as much public attention [3]. Group 2: Evaluation Process - An evaluation task was conducted to test the visual recognition capabilities of various AI tools, inspired by a recent international AI competition [5][12]. - The evaluation involved ten images of confusing restroom signs, with a scoring system based on correct identification [11][15]. - Zhipu's GLM-4.5 model (without reasoning) scored the highest at 86 points, while the reasoning version and ChatGPT's GPT-5 both scored 78 points [12]. Group 3: Performance Insights - The evaluation revealed that Zhipu's models made errors in identifying restroom signs, with the non-reasoning version being the only one to answer incorrectly on one of the ten questions [26]. - Other AI tools, such as Doubao and Kimi, performed better in certain instances, showcasing the varying capabilities of different models [26][23]. - The evaluation highlighted the potential for AI tools to improve in visual recognition tasks, which could have significant applications across various industries [39][42].

省流版:当初用电容是因为视觉检测能力不够强，够强后就决定取消电容了，理想认为因为更强的视 觉检测可以适应更高的速度域，实际体验较过去的电容版本更强了。 以下为理想第一产品线负责人老汤哥原文： 看样子又可以来写个小故事了，写一写理想汽车方向盘监测的进化史了。 最早在2019年的理想ONE的时代，这是我们开发的第一款带辅助驾驶的车型，当时方向盘的监测到 底用什么方式就经历过了讨论，（其实方向盘监测是法规要求，并不是一个功能配置）。 当时技术路线有2条： 第一条，特斯拉的扭矩方案，方向盘需要轻轻的掰一下，给一个相对还不小的力。 第二条，用电容，说是电容，其实很多时候需要 手"捏"一下方向盘，因为需要比较大的"手"接触面 积。 最后大家讨论下来，还是选择了电容方式（其实扭矩也具备，只是主要用电容）。 因为当时特斯拉的体验并不好，需要用力的"掰"一下方向盘。 其实我们当时用这个方式，还是蛮大的挑战，因为和特斯拉的方式不太一样，和行业常规方案都不太 一样。 不过还好，可能也因为当时理想ONE的辅助驾驶就是一个ME的外包方案，也不是行业的头部，也没 有什么风浪。 到了理想ONE的2021款，我们辅助驾驶切换成了自研的方案， ...