Mobileye's $900M Bet on a Robot That Learns From One Video

Mentee Robotics built a humanoid robot that could learn a new warehouse task from a single video demonstration, without teleoperation, without engineering involvement, and with near-100% manipulation success rates in internal testing. In January 2026, Mobileye agreed to acquire the company for 00 million, combining Mobileye's autonomous driving perception stack with Mentee's physical AI and humanoid locomotion expertise.

Founded in 2021 in Herzliya, Israel, by Amnon Shashua (Chairman, also CEO of Mobileye), Professor Lior Wolf (CEO), and Professor Shai Shalev-Shwartz, Mentee raised 0 million across multiple rounds before its acquisition. The MenteeBot V3 demonstrated two robots collaboratively sorting 32 boxes in an uncut 18-minute warehouse video in November 2025, one of the most compelling real-world humanoid capability demonstrations published by any company that year.

AI-generated image

MenteeBot's hands integrate motor-based tactile sensing, enabling reliable grasping of rigid objects up to 25 kg. Source: AI visualization

The Israel AI Lab That Built a Robot You Can Mentor

Mentee Robotics emerged from a concentration of computer vision and machine learning expertise that is unusual even by Israeli tech standards. Amnon Shashua co-founded Mobileye in 1999 and built it into the dominant supplier of vision-based driver assistance systems before Intel acquired it for 5.3 billion in 2017. He brought that pattern recognition expertise, plus deep relationships with semiconductor and compute suppliers, to Mentee Robotics when he co-founded it in 2021.

Lior Wolf, serving as CEO, brings academic credentials in deep learning and computer vision from Tel Aviv University. Shai Shalev-Shwartz, the third co-founder, is known for foundational work in machine learning optimization and reinforcement learning. The founding team's combined academic and industry track record is what attracted investors like Cisco Ventures, Samsung, Ahren Innovation Capital, and 10DVC in a market where most robotics investors were backing companies with stronger hardware backgrounds and lighter AI depth.

The company name encodes the core philosophy. "Mentee" refers to the robot as a learner, something that can be mentored by a human using natural methods: demonstrating a task once, giving verbal instructions, or showing it a short video. The contrast with conventional robot programming, which requires weeks of engineering effort per task, is the product's primary selling point.

Under the Hood: The MenteeBot AI Architecture

MenteeBot's technical stack is built around three integrated systems. The perception layer uses camera-only sensing combined with NeRF (Neural Radiance Field) and 3D Gaussian Splatting (3DGS) for 3D scene mapping, OWLv2 and nano-SAM for object detection and segmentation, and visual odometry for localization. There is no LiDAR and no expensive depth sensor array. This is a deliberate design choice: cameras are cheap, reliable, and produce exactly the kind of visual data that large vision models were trained on.

The planning layer uses a reinforcement learning-based control system trained primarily in simulation, with Sim2Real transfer handling the gap between simulated and real environments. The system implements automatic curriculum learning, progressively increasing task difficulty during training to build robustness across lighting variations, object placement variations, and environmental noise. The result is manipulation and locomotion behaviors that generalize to new environments without per-environment retraining.

The task reasoning layer is where Mentee's AI-first approach most clearly distinguishes the MenteeBot from competitors. Rather than programming tasks as fixed sequences of joint angles, the system decomposes natural language or demonstration inputs into a code-like flow with explicit error handling, structured like a try-catch block in software. If a grasping attempt fails, the robot retries with a modified approach rather than stopping and waiting for human intervention. This architecture enables the near-100% task completion rates Mentee reported in its November 2025 warehouse demonstration.

Two MenteeBot V3 units collaboratively sorting 32 boxes in Mentee Robotics' November 2025 warehouse demonstration video. The 18-minute uncut video was widely cited as one of the most credible real-world humanoid capability proofs of 2025. Source: AI visualization

The 00 Million Acquisition: Why Mobileye Wants a Humanoid

The Mobileye-Mentee deal, announced at CES on January 6, 2026, is structured as 12 million in cash plus approximately 26.2 million Mobileye shares, for a total consideration of roughly 00 million. The deal was expected to close in Q1 2026. For a company that raised 0 million total across five years, the acquisition multiple is exceptional.

The strategic logic from Mobileye's perspective is about extending its perception stack from vehicles to robots. Mobileye's core product is a camera-based perception system that processes visual input from a vehicle's cameras to identify lanes, pedestrians, cyclists, and other road hazards. That is, at a fundamental level, the same problem MenteeBot solves: understanding a physical environment through cameras and taking reliable physical actions in response.

Amnon Shashua has described the acquisition as central to Mobileye's broader "Physical AI" strategy, expanding from autonomous vehicles to any physical system that needs to perceive and act in the real world. The term Physical AI is being used across the robotics industry to describe AI systems that operate on physical hardware rather than processing text or images passively, and Mobileye is betting that its perception expertise transfers directly to humanoid robotics.

What the Acquisition Means for the Humanoid Market

The Mentee acquisition is the first major M&A transaction in the humanoid robot sector at scale, and it signals something important: the competition for humanoid robotics talent and technology has moved beyond startup funding rounds into acquisitions by large-cap technology and automotive companies. Mobileye paid roughly 22 times Mentee's total funding raised to acquire the company, a premium that reflects both the difficulty of building what Mentee built and the strategic value of owning it before competitors do.

The implications for the broader humanoid market are significant. Other automotive and semiconductor companies are watching. Qualcomm, NVIDIA, and Infineon all have existing relationships with robotics startups. Intel (Mobileye's parent company) now has a direct humanoid robot asset for the first time. Toyota, which has invested heavily in its own humanoid research program for decades, may face competitive pressure from Mobileye-backed deployments. The acquisition compresses the timeline for credible humanoid deployment in automotive manufacturing, which happens to be the industry BYD, Tesla, and every major automaker is simultaneously expanding.

For smaller humanoid startups, the Mentee deal is a double-edged signal. On one hand, it validates the sector and establishes acquisition as a plausible exit path. On the other hand, it removes one of the more technically credible independent players from the competitive field and places it under a company with deep manufacturing and distribution relationships in the automotive sector.

What's Coming Next: Post-Acquisition Integration

The immediate question after the acquisition closes is whether Mobileye deploys MenteeBot technology through its existing automotive OEM relationships. Mobileye has direct contracts with more than 50 automotive manufacturers worldwide. If even a fraction of those relationships extend to humanoid robot procurement for factory floor applications, the deployment scale for MenteeBot could ramp faster than any independent startup could achieve on its own capital.

The MenteeBot V3.1 was teased in February 2026, with improved industrial robustness characteristics. The development roadmap under Mobileye ownership will likely prioritize automotive manufacturing environments, where the structured, repetitive nature of assembly tasks aligns well with current humanoid capabilities, over the home robotics applications Mentee had also been exploring. Expect to see Mobileye announce automotive factory pilot programs for MenteeBot in 2026-2027.

Frequently Asked Questions

The 12-Month Outlook

The Mobileye-Mentee combination will spend 2026 integrating Mentee's AI stack with Mobileye's existing automotive customer base and chip architecture. Watch for: a formal MenteeBot commercial product announcement under the Mobileye brand, pilot program announcements with one or more automotive OEMs, and potential integration of MenteeBot's perception approach into Mobileye's broader autonomous vehicle sensor suite.

The acquisition also accelerates the timeline for other automotive and semiconductor companies to make similar moves. NVIDIA's investment in multiple humanoid startups, Qualcomm's robotics platform push, and Toyota's in-house humanoid development program all represent competing bets on who controls the physical AI layer in manufacturing automation. The first company to deploy 1,000 humanoid robots in a single automotive factory will establish the credibility benchmark the entire sector is racing toward.

The humanoid robot market is bifurcating: large-funded independents like Figure AI and Apptronik pursue OEM customer contracts on their own timelines, while AI-first specialists like Mentee become acquisition targets for companies with deep manufacturing distribution and capital to deploy at scale. Both paths lead to the factory floor. The question is which gets there first.