Sanctuary AI — hydraulic humanoid robots with human-like hands

• 2018
  Founded
• Vancouver, BC
  Headquarters
• Phoenix Gen 7
  Latest flagship (2024)
• $140M+ raised
  Funding total
• Canada
  Country
• Private
  Status

About Sanctuary AI

Sanctuary AI — legally Sanctuary Cognitive Systems Corporation — was founded in Vancouver, British Columbia in 2018 by Geordie Rose, Suzanne Gildert, Ajay Agrawal, and Olivia Norton. Rose is best known as the co-founder of quantum-computing pioneer D-Wave Systems and, later, the AI startup Kindred; Gildert previously led AI research at Kindred. From the outset, Sanctuary framed its mission in unusually ambitious terms: to create the world's first human-like intelligence in a general-purpose robot, meaning a single machine that can perform the wide range of physical tasks a human worker can — not a bespoke industrial arm tuned to one operation.

The company's early progress was iterative and hardware-first. Sanctuary's fifth-generation robot ran a widely covered week-long retail pilot at a Mark's store — a Canadian Tire subsidiary — in Langley, British Columbia in January 2023, completing 110 distinct retail tasks under real store conditions. In April 2024, Sanctuary unveiled the seventh generation of its humanoid — the version widely referred to as Phoenix Gen 7 — with further miniaturized hydraulic actuators, expanded range of motion, and improved hand durability. The same month, Sanctuary announced a strategic partnership with automotive giant Magna International that combined an equity investment with a pilot deployment of Phoenix inside Magna's manufacturing operations. In May 2024, a Microsoft collaboration added Azure cloud infrastructure and joint AI research to Sanctuary's technology stack, positioning the company as one of the more strategically embedded humanoid programs in the industry.

On the capital side, Sanctuary closed an oversubscribed C$75.5 million (~US$58.5 million) Series A in March 2022 with a corporate-heavy investor syndicate: Bell, Magna, Verizon Ventures, Workday Ventures, Evok Innovations, Export Development Canada, SE Health, and Accenture. Later in 2022, the Government of Canada added a C$30 million Strategic Innovation Fund contribution — one of the largest federal AI commitments made to a single company at the time — and additional strategic financing followed from BDC Capital, British Columbia's InBC Investment Corp, and Microsoft, bringing publicly reported cumulative funding above C$140 million by mid-2024.

The competitive landscape has intensified sharply since Sanctuary's founding. Where general-purpose humanoids were a niche research bet in 2018, they are now a headline-grabbing category driven by the current AI boom, with well-funded rivals including Figure AI, Apptronik, 1X, Tesla Optimus, Agility Robotics, and Boston Dynamics all pursuing broadly overlapping visions. Sanctuary's core differentiator is its bet on hydraulic dexterity — its 20-DoF hydraulic hands with millinewton-scale tactile sensitivity remain among the most dexterous end-effectors demonstrated on any humanoid platform. Both original co-founders, Rose and Gildert, departed the company in November 2024, and Sanctuary has since publicly emphasized a broader Physical AI strategy — deploying its Carbon control stack on existing industrial robots alongside the Phoenix platform itself.

Technology stack

Sanctuary's platform is a tightly integrated combination of custom hydraulic mechatronics, multi-modal perception, human-scale dexterous manipulation, and the in-house Carbon AI control system. Where most competing humanoid programs have converged on electric brushless-DC actuation, Sanctuary is one of the few in the industry to publicly commit to miniaturized hydraulic actuation throughout the manipulators — a significant technical distinction with real implications for force density, precision, and component supply chain.

Perception

Phoenix robots are equipped with embedded micro-cameras for vision and an unusually rich tactile-sensing array at the manipulators. Published information describes tactile pads with roughly seven force-sensing cells per finger and micro-barometer-based sensing capable of resolving contact forces down to approximately 5 millinewtons, within about 40% of the sensitivity of a human fingertip. This combination of high-resolution touch with force-torque sensing is what allows Phoenix to reliably grip deformable objects, insert connectors, and handle delicate items in unstructured environments — the kind of tasks that plain vision-guided arms typically struggle with.

Carbon AI control system

Carbon is Sanctuary's in-house control architecture — described by the company as a brain-like system that blends learned behavior with symbolic reasoning. In practice, human operators teleoperate Phoenix through motion-capture rigs to demonstrate a task; Carbon converts those demonstrations into training data and layers in autonomy over time. Following the May 2024 collaboration with Microsoft, Sanctuary trains and runs large portions of the Carbon stack on Microsoft Azure, with the company publicly positioning Carbon as its route to what it terms Large Behavior Models — the physical-world counterpart to large language models.

Actuators, power, and compute

The most technically distinctive element of Phoenix is its proprietary miniaturized hydraulic valves, developed in-house, which drive the robot's dexterous end-effectors. Hydraulics deliver very high force density in a small envelope — useful for a 25 kg payload rating and for the fine-grained forces required in contact-rich manipulation — at the cost of added mechanical complexity relative to pure electric actuation. Phoenix pairs those hydraulic hands with an electric mobility base and an onboard battery targeting roughly four hours of continuous operation, with onboard compute sized for the Carbon inference stack while heavier training runs on Azure infrastructure.

The electronics burden of a humanoid platform like Phoenix is substantial. Each hydraulic valve requires precision drive electronics and analog conditioning; every tactile sensor cell needs low-noise front-end amplification and multi-channel data acquisition; the electric mobility base and joint controllers benefit heavily from wide-bandgap semiconductors — SiC and GaN devices in the motor-driver and power-stage sections — for higher switching efficiency and lower thermal load. Precision passives, MLCC capacitors across every rail, and battery-management ICs with cell balancing, thermal monitoring, and multi-stage protection are all first-class subsystems whose reliability directly determines how long any humanoid platform can operate between charges.

The robotics supply chain — where Hybrid Electronics fits in

A modern humanoid robot is one of the most component-dense products ever built. A single Phoenix-class platform depends on thousands of individual electronic components: precision BLDC motor drivers for the mobility base and non-hydraulic joints, high-torque servo modules, IMUs and 6-DoF force-torque sensors, LiDAR and time-of-flight modules, GPU compute boards for Carbon inference at the edge, battery-management ICs, wide-bandgap SiC/GaN power semiconductors, dedicated valve-drive electronics for each hydraulic actuator, high-speed board-to-board connectors, precision resistors, and thousands of MLCC capacitors distributed across every power and signal rail.

Building that kind of hardware puts enormous pressure on the electronic-components supply chain. Prototype pilots and low-volume production runs frequently hit allocation, EOL notices, and long lead times on the exact parts they need, and R&D and field-service teams routinely have to support older-generation platforms that use semiconductors no longer in mainstream distribution. During the current humanoid robotics and general-purpose robot AI boom, this pressure is amplified: dozens of well-funded teams — Sanctuary, Figure AI, Apptronik, 1X, Tesla Optimus, Agility Robotics, and Boston Dynamics among them — are chasing the same short list of high-torque BLDC drivers, precision IMUs, GPU compute modules, and automotive-grade power semiconductors, and franchise stock on many key parts is now measured in months of lead time rather than weeks.

Hybrid Electronics has been sourcing electronic components for industrial and embedded systems since 1995. For robotics teams — humanoid, mobile, industrial, and research — we specialize in finding the parts you need when mainstream distribution can't:

Trademark and affiliation notice. Sanctuary AI®, Phoenix, and Carbon are trademarks of Sanctuary Cognitive Systems Corporation. Hybrid Electronics is an independent electronic-components sourcing company and is not affiliated with, endorsed by, or authorized by Sanctuary Cognitive Systems Corporation. All information on this page is compiled from public sources for educational and reference purposes.