苹果早在2019年就启动这款机器人项目，内部代号J595。项目初期定位保守，主要设想用于视频通话；机械臂的作用是跟随用户在房间内移动。但随着管理层希望为它叠加AI能力，项目目标大幅膨胀。
按照当前设计方案，这款机器人搭载摄像头与各类传感器，可以完整感知周边环境、自主行动：当有人走进房间，它可以开灯，甚至尝试驱赶桌上的猫咪；同时内置避障能力，防止磕碰周围物品。
参与该项目的人士透露，机器人大部分硬件方案大约两年前就已经定型，但软件团队一直难以跟上进度。负责J595的硬件团队希望加快节奏、尽快发布产品；而软件团队主张推迟上市。过去一年里，软件团队一直在精简原定功能，打算把产品核心能力收敛到FaceTime通话，以及开灯这类简单家庭自动化任务。
苹果计划这款机器人售价约1000美元，受元器件行情影响，最终售价可能更高。多名前员工称，它或将命名为Home View Ultra，沿用上月苹果发布的固定式桌面设备Home View的命名体系。苹果原本希望明年推出，但发布时间也有可能推迟至2028年（硬件团队最初计划今年就上市）。
这款机器人的研发困境，折射出特纳斯上任初期苹果面临的挑战。苹果擅长每年对iPhone、Mac做迭代式更新，但并不擅长从零开发全新品类，公司一贯节奏偏平缓。例如Vision Pro头显前后打磨约8年，两年前正式推出，市场反响平淡。如果J595明年面世，研发周期也将接近这个时长。
在AI浪潮下，苹果的步伐尤其迟缓。Siri的重大改版耗时数年，即便今年上线全新AI功能，在AI智能体赛道上，它依旧落后于Meta的Muse、OpenAI的Dots等竞品。
前任CEO蒂姆・库克为配合特纳斯9月履新，推迟了多款新品的发布，让新品亮相的时间点和他上任时间同步。此后苹果陆续发布Home View、折叠iPhoneDuo以及触控屏MacBook，但市场评价褒贬不一：2899美元的MacBook被诟病定价过高；Duo热度很高，却频现屏幕碎裂投诉；Home View则被批评只是放大版ApplWatch，高度依赖表现依旧不尽人意的Siri。
特纳斯接手之时，一众强劲对手都在布局AI硬件与机器人。一年前，OpenAI收购了由苹果前首席设计官乔尼・艾夫联合创办的硬件初创公司。此后艾夫全心为OpenAI打造AI硬件产品线，OpenAI还挖走400多名苹果工程师投入相关项目。OpenAI计划明年推出一款甜甜圈造型智能音箱，自带可动结构与摄像头，感知用户和周边环境。
Meta也密集推出硬件，包括AI眼镜、VR头显。相比苹果，Meta部分项目从概念到落地速度很快：其口袋AI设备Charm据称研发不到一年，预计12月出货。
特斯拉正在研发人形机器人Optimus，马斯克称它将协助处理家务。大量初创企业也加入竞争，比如机器人公司1X，原本计划今年把人形机器人送入家庭。它的Neo机器人去年一经曝光就走红，视频展示它洗碗、叠衣服。
除J595之外，在资深高管凯文・林奇带领下，苹果机器人团队还探索其他形态：轮式跟随机器人、室内飞行机器人。苹果还有一个项目，研发专门完成特定日常家务的机器人，比如叠衣机器人。理论上，只聚焦少数固定家务的机器人，开发难度远低于需要处理多样任务的通用人形机器人。知情会议记录显示，特纳斯对家务专用机器人方向兴趣浓厚。
一名参与过J595项目的前工程师这样评价：苹果“为赶上AI浪潮，已经被折腾得四分五裂”。
“棍子上的iPad”
J595的研发历时近十年，先后由多名苹果高管接手，不少负责人同时兼顾多个项目，很难全身心投入这款设备。彭博此前已经披露过该机器人的部分信息。
最早的原型机在2018年完成。次年库克观摩原型演示，批准项目继续推进。早期原型机尺寸比现在的方案更大，一些苹果工程师调侃它，看上去就像“一根棍子上插着一台iPad”。
机器人硬件团队由马修・科斯特洛牵头，他此前负责Beats（苹果2014年收购的耳机品牌）；软件开发交给苹果特殊项目组，当时这个部门主要在推进泰坦计划——苹果自动驾驶汽车项目。有很长一段时间，资深工程师布莱恩・林奇（iPod时代老员工）同时负责汽车座舱和J595：他设想这款机器人不只放在家里，还可以安置在苹果汽车的中控台上。
2021年，特殊项目组负责人更替：道格・菲尔德离职转投福特，凯文・林奇接任。林奇因给Apple Watch开发健康追踪功能、挽救这条产品线而闻名，但当时他大部分精力放在自动驾驶汽车上，机器人业务主要由克里斯・萨里负责，萨里曾主导Apple Watch软件开发。
早在早期阶段，项目成员就清楚，如果要支撑高昂定价，它不能只是一台花哨的FaceTime通话设备。2022年ChatGPT问世掀起AI热潮之后，团队开始增加AI能力，打造更复杂的体验。团队内部希望加入大量主动式智能体功能，让设备自主完成多样任务。机器人部门还专门设立智能体AI研究分支，由弗拉登・科尔图恩带领；他曾在英特尔负责AI业务，2021年加入苹果。
但苹果自研AI能力，和竞品差距明显。因此团队获准使用外部可用模型，包括开源模型、谷歌授权模型。团队原本计划后续将J595和Siri打通，但在机器人开发早期，Siri团队和机器人团队相互独立，Siri团队本身的迭代压力已经很大。
2024年苹果叫停自动驾驶汽车项目后，林奇将全部重心转向家用机器人，大量原汽车项目员工被调配过来。三名项目相关人士表示，不少更早就在做机器人的员工对他的管理方式感到失望，认为他很少沟通、听取一线意见。
林奇选择下调初代J595的功能预期：计划先推出基础版本，后续通过更新逐步增加能力，思路和当年Apple Watch一致。初代Apple Watch 2015年发布，后续才加入健康健身传感器，才让这款穿戴设备真正走红。
特纳斯接手掌舵
在升任CEO之前，时任硬件负责人的特纳斯对很多新项目持怀疑态度，比如夭折的自动驾驶项目、Vision Pro头显。但三名熟悉机器人项目的人士称，他对苹果机器人赛道抱有更高期待。
去年，凯文・林奇以及特殊项目组的汇报线，从高级副总裁约翰・詹南德雷亚划归特纳斯（詹南德雷亚曾主管苹果AI业务，已于今年早些时候离职）。
4月苹果官宣特纳斯将于年内接任CEO。即便还在交接准备阶段，他几乎出席每一场J595和其他机器人的演示会、产品规划会。
知情人士介绍，特纳斯在全公司推行策略：精简团队、加快落地。机器人团队规模目前依旧偏小，苹果深耕机器人领域的专业人才稀缺，有人估算真正的机器人专家仅10人左右。林奇麾下软件团队约300人，其中30人左右负责研究。
J595项目成员已经在讨论第二代机型，打算把这一代砍掉的部分功能放到下一代。与此同时，项目早期一批工程师已经离职：硬件负责人布莱恩・林奇去往健身硬件厂商Oura，机器人研究负责人张健跳槽Meta。
一部分J595团队成员希望产品尽快上市；也有人担心，它会遭遇和Home View一样冷淡的市场反响。但他们顾虑，苹果拖得越晚发布，行业AI能力持续升级，外界对这款机器人的期待值也会水涨船高。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Apple started this robot project in 2019, internal code J595. The initial positioning of the project is conservative and is primarily envisaged for video calls; the role of the arm is to follow the user within the room. However, as management wished to superimpose AI capabilities for it, the project objectives expanded significantly.
According to the current design, this robot carries a camera and a variety of sensors that provide a complete sense of the surrounding environment and autonomous action: When someone enters the room, he can turn on the light and even try to drive out the cats on the table; at the same time, he has an internal shield to prevent contact with the surrounding objects.
Participants in the project revealed that most of the hardware programmes for robots had been stereotyped about two years ago, but the software team had been unable to keep pace. The hardware team in charge of J595 hopes to speed up the pace and speed of the publish product, while the software team advocates late listing. During the past year, the software team has been streamlining its original functionality, with the intention of constricting core product competencies to FaceTime calls and simple home automation tasks such as lighting.
Apple Project, a robot with a price of about $1,000, is likely to end up at a much higher price, influenced by metaware conditions. Several former employees stated that it might be named Home View Ultra, following the name system of Home View, the fixed desktop device of Apple publish last month. Apple was expected to be launched next year, but publish may also be delayed until 2028 (the hardware team initially planned to be listed this year).
This robot's development dilemma mirrors the challenges of the early apples of Turner. Apples are good at an iterative update of iPhone and Mac each year, but they are not very good at developing entirely new products from zero, and companies have been slow and flat. For example, the Vision Pro has been on display for about eight years, officially launched two years ago, and the market has been quiet. If J595 is born next year, the R & D cycle will be close to that.
Apples are particularly slow in the waves. Siri’s major adaptation took several years, and even with the new AI functionality on line this year, it still lags behind the races of Muse in Meta and Dots in Openai.
The former CEO Tim Cook postponed publish, a number of new items, to coincide with his arrival in September. Since then, apples have continued with publish Zhome View, the folding of iPhoneDuo and touchscreen MacBook, but market evaluations have been mixed: $2899 MacBook is overpriced; Duo is very hot, but the screens are broken; Home View has been criticized for simply scaling up the Applewatch, with high reliance on Siri, whose performance remains unsatisfactory.
When Turners took over, a lot of powerful rivals were putting together AI hardware and robots. One year ago, Open AI acquired the hardware start-up company, a joint venture of former Apple Chief Design Officer Johnny Ive. Since then, Eiff has devoted all his efforts to developing the AI hardware product line for Open AI, which has also dug up more than 400 apple engineers for the project. Open AI plans to launch next year a donut-constructed smartphone with self-portable structures and cameras, a sense of users and surrounding environments.
Meta is also intensely rolling out hardware, including Ai glasses and VR headlines. Compared to apples, some of the Meta projects are moving from concept to landing: their pocket AI equipment, Charm, was allegedly developed less than a year ago and is expected to be shipped in December.
Tesla is developing Optimus, a human robot, which Mask says will assist with housework. A large number of start-ups are also competing, such as the robot company 1X, which was supposed to send human robots to their homes this year. Its Neo robot went red when it was exposed last year, and the video shows it washing dishes and folding clothes.
In addition to J595 and led by senior executive Kevin Lynch, the Apple Robot team explored other forms: wheeled robots, indoor flying robots. Apple also has a project to develop robots specializing in certain routine household chores, such as coating robots. Theoretically, only a small number of stationary domestic robots are less developed than universal human-shaped robots that need to deal with a variety of tasks. The records of the informed meetings show that there is a strong interest in the direction of domestic robots.
A former engineer who had participated in the J595 project assessed the apple as having been “disturbed to catch up with the AI wave”.
"iPad on a stick"
The development of J595 has lasted for almost 10 years and has been taken over by a number of apple executives, many of whom have a combination of projects, making it difficult for them to fully invest in the equipment. Bloomberg has previously disclosed some of the robot's information.
The earliest prototype was completed in 2018. The following year, Cook observed the prototype demonstration and approved the project for continuation. The early prototype was larger than the current one, and some apple engineers turned it like “an iPad on a stick”.
The robotic hardware team was led by Matthew Costello, who was previously in charge of Beats (the Apple headphone brand acquired in 2014), and the software development was handed over to the Apple Special Project Team, which was mainly promoting the Titan project, the Apple Autopilot. For a long time, a senior engineer, Brian Lynch, was in charge of both the car cabin and J595: He assumed that the robot was not just at home, but at the center of the apple car.
In 2021, the head of the special project team was replaced: Doug Field was transferred to Ford and Kevin Lynch took over. Lynch was known for developing a health tracking feature for Apple Watch and saving the product line, but at the time he was mostly focused on auto-driving cars, where the robotics business was primarily carried out by Chris Surly, who had led Apple Watch software development.
At an early stage, it became clear to the project members that, if high pricing was to be sustained, it could not be merely a fancy Facetime telephone device. In 2022, after the advent of ChatGPT, the team began to increase AI capacity and build more complex experiences. Within the team, a large number of active AI agent features are expected to be added to enable the equipment to perform its diverse tasks autonomously. The robotics sector has also set up the AI agent AI research branch, headed by Vladon Cortoun, who was in charge of AI operations in Intel and joined Apple in 2021.
But there is a clear gap between apple self-study and competition. The team was therefore allowed to use externally available models, including open source models and Google Authorization models. The team had planned to follow up on J595 and Siri, but in the early stages of robot development, the Siri team and the robot team were independent of each other and the team itself was already under considerable stress.
In 2024, after the Apple called for the Autopilot project, Lynch shifted the entire focus to domestic robots, and a large number of original car project staff were deployed. Three project-related sources indicated that many of the earlier robotics staff were disappointed with his management and found that he rarely communicated and heard first-line views.
Lynch chose to downplay the functional expectations of the original J595: the plan was to launch the base version and to build up capacity by updating, along the lines of Apple Watch of the year. The original Apple Watch 2015 publish was later added to the Healthy Fitness Sensor to make the wearing of the device truly red.
Turner takes the wheel.
Before being promoted to CEO, the then hardware manager, Turner, had doubts about many new projects, such as the premature autopilot project and Vision Pro. However, three people familiar with the robotics project stated that he had higher expectations of the apple robot track.
Last year, Kevin Lynch and the Special Project Team reporting lines were transferred from John Janandreya, Senior Vice-President, to Tennessee (Janandreya was in charge of AppleAi operations and left earlier this year).
Apple official Tanners will succeed CEO in April. He attended almost every J595 and other robot demonstration, product planning, even in the pre-shipment phase.
According to the source, Turners is pursuing a corporate strategy: streamlining teams and accelerating landings. The size of the robot team is still small, and there is a scarcity of expertise in the field of apple-dry robots, with some estimating that there are only about 10 real robot experts. The Lynch software team is about 300 people, of whom about 30 are in charge of research.
Project J595 members are already discussing a second-generation model, with the intention of placing some of the cut-off functions of this generation in the next generation. At the same time, an early group of engineers of the project had left: the hardware manager, Brian Lynch, went to the fitness hardware manufacturer, Laura, and the robot research manager, Zhang Kenti, Meta.
Some members of the J595 team wanted the product to be on the market as soon as possible; there were also concerns that it would suffer as cold a market reaction as Home View. However, they were concerned that the later the apples were dragged to publish, that the industry's AI capabilities were constantly upgraded and that the expectations of the robot would rise.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.