Machine Learning Research Intern - Reinforcement/Imitation Learning

Sanctuary is on a mission to create the world's first human-like intelligence in general-purpose robots that will help us work more safely, efficiently, and sustainably, helping to address the labour challenges facing many organizations today.

Sanctuary AI will sell its robot brain, not just its Phoenix humanoid. Sanctuary AI hit a 99.5% success rate on a factory wire-plugging task using its Carbon AI system on a rival's robotic arm, not its own Phoenix humanoid. The Vancouver company, now led by CEO Daniel Friedmann, is betting its future on licensing that AI brain rather than out-building Figure AI and Tesla on hardware. Sanctuary AI has spent years selling Phoenix as the body for its robot intelligence. Its latest factory proof says Carbon may travel faster when the full humanoid stays out of the way. Sanctuary AI just told the humanoid robotics industry something uncomfortable: its intelligence no longer needs its own robot attached. The proof came from a factory floor. The Vancouver company's AI control system, called Carbon, hit a 99.5% task success rate on a wire-plugging job for a global Tier 1 automotive supplier, threading flexible wires into moving targets at a 2.54-second cycle time. Sanctuary said on June 17 that the result was validated against the customer's live production benchmarks, and The Robot Report also covered the industrial robotics shift. The important detail is plain enough. This was not a Phoenix humanoid walking up to a line and doing a person-shaped job. That is the pivot. Sanctuary unveiled Phoenix as its sixth-generation general-purpose robot in May 2023, then announced a seventh generation in April 2024 and a Phoenix Generation 8 platform built for higher-quality data capture after that. It built the body and the sensors, the hands and the control software, all as one system. Now, under CEO Daniel Friedmann, who Sanctuary says moved from chairman to chief executive in June after nearly two decades running aerospace company MDA, the company is pushing Carbon as a software layer for existing industrial robots. If you're watching this market, that matters more than another glossy humanoid demo. Frankly, it looks like an admission that hardware is not where Sanctuary can win fastest. That's the tell. Outgunned on hardware, so it is selling the brain. Figure AI closed more than $1 billion in Series C funding in September 2025 at a $39 billion post-money valuation, according to Figure's own announcement. It is scaling Helix, its embodied intelligence system, and BotQ, a factory whose first line was designed for up to 12,000 humanoids a year. Figure also said its supply chain could scale to 100,000 robots over four years. That's the scale Sanctuary is up against. Tesla is trying to make Optimus part of a much larger manufacturing machine, even as its 2026 volume-production language has shifted. Unitree is pressing from the other side, with its own store listing the G1 at $13,500 and the R1 from $4,900. Unitree Robotics opened book-building for its Shanghai STAR Market IPO on August 5, targeting a valuation near 42 billion yuan, just days after new US sanctions hit the company. It's the first profitable, at-scale humanoid robot maker to go public, setting a real valuation benchmark against pre-revenue rivals like Figure AI. - Unitree robotics Shanghai IPO valuation - humanoid robot company IPO pricing Sanctuary is playing a different hand. Bloomberg reported in late 2024 that the company had raised C$140 million as of early July that year, and CB Insights now lists total funding at about $148.6 million. That is real money, but it is not Figure money, and it is not Tesla money. You do not beat that field by pretending capital intensity is a detail. You find the layer where a smaller company can still matter. For Sanctuary, that layer is Carbon. The company already had a clue in its own language. When Phoenix was announced in 2023, Sanctuary described Carbon as the AI control system that gave the robot its ability to act, reason, plan. Its 2024 Microsoft collaboration said Sanctuary would use Azure infrastructure for training and inference, and for networking and storage, tied to Carbon and its Phoenix robots. The body got the photos. The brain did the compounding. This is not a new idea in robotics. Nvidia's Isaac GR00T N1.7 is an open vision-language-action model for generalized humanoid skills, with model weights and reference code available through GitHub and Hugging Face. Nvidia's developer blog describes GR00T 1.7 as commercially usable and released under Apache 2.0. Same playbook, different company. The smartphone industry already taught you the broader point: the operating system can capture value even when someone else makes the device. Sanctuary's version has a wrinkle Nvidia does not have to deal with. Nvidia does not sell a competing humanoid body. Sanctuary does. If a factory can get near-production reliability from Carbon running on a robot arm it already understands, Phoenix becomes less urgent as the first sale. Sanctuary's answer, for now, seems to be that it can live with that. Licensing Carbon on installed hardware should close faster than selling a full humanoid, and every deployment gives Sanctuary more task data for the one asset it is clearly betting on. The bet is blunt: the moat may not be the body at all. What still needs to be proven. Figure, Tesla, 1X: all three are still building around full-stack humanoids, where hardware and software improve together under one roof. 1X, for example, said in July that its NEO hands are tightly integrated with in-house motors and electronics, tendon systems and firmware. Sanctuary's wire-plugging result challenges that instinct. A contact-rich task, done at line speed, without the full Phoenix form doing the work, is exactly the kind of evidence the humanoid market tends to skip past when it is busy watching robots fold laundry or wave onstage. Still, the missing details matter. Sanctuary has not named the automotive supplier, and it has not published pricing for Carbon as a separate product from Phoenix. It also has not shown that the same 99.5% performance carries across a long list of customer sites, parts and robot platforms. A wire harness on one line is a serious proof point. It is not a business model by itself. SoftBank is in early talks to anchor an $800 million funding round for Agile Robots, the Munich-based robotics company in which the Vision Fund 2 previously led a $220 million Series C in 2021. The potential deal would rank among the largest single raises in physical AI hardware and arrives as SoftBank executes a sweeping pivot toward robotics... - SoftBank 800 million humanoid robot funding round - physical AI robotics companies raising capital investment Watch for two things next: a customer willing to put its name on the deployment, and a price that makes Carbon cheaper to adopt than waiting for humanoid hardware to mature. Those will say more about Sanctuary's pivot than any cycle-time statistic. Until then, the company has at least made one point clearly. If the robot brain can move from body to body, the race for humanoids is no longer only about who builds the best machine. Elroy is a digital marketer and developer from Goa, with over a decade of experience web development and marketing. He has been associated with several startups and serves currently as an Editor to the Asia Pacific Industrial magazine. He occasionally writes on Startup Fortune about technology and automation.