Phoenix (Sanctuary AI)
Last reviewed
May 17, 2026
Sources
30 citations
Review status
Source-backed
Revision
v5 · 5,274 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 17, 2026
Sources
30 citations
Review status
Source-backed
Revision
v5 · 5,274 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Phoenix | |
|---|---|
 | |
| General information | |
| Manufacturer | Sanctuary AI |
| Type | General-purpose humanoid robot |
| Country of origin | Canada |
| Year introduced | 2023 (Generation 6) |
| Current generation | Generation 8 (December 2024) |
| Status | Pilot deployment |
| AI control system | Carbon |
| Website | sanctuary.ai |
Phoenix is a general-purpose humanoid robot developed by Sanctuary AI, a robotics and artificial intelligence company headquartered in Vancouver, British Columbia, Canada. Powered by Carbon, Sanctuary's proprietary AI control system, Phoenix is designed to perform a wide range of work tasks across industries including manufacturing, retail, and logistics. The robot is notable for its highly dexterous hydraulic hands, which feature 21 degrees of freedom per hand, and for its teleoperation-to-autonomy training pipeline that uses human-guided data collection to progressively build autonomous capabilities.
First publicly unveiled in May 2023 as the sixth-generation version of Sanctuary's robot platform, Phoenix has undergone rapid iteration, reaching its eighth generation by December 2024. The robot has been commercially piloted at Mark's (a Canadian Tire subsidiary) retail stores and is the subject of a strategic manufacturing partnership with Magna International, one of the world's largest automotive parts suppliers. As of early 2026, Sanctuary AI has raised over $147 million in total funding.[1][2]
Sanctuary AI (formally Sanctuary Cognitive Systems Corporation) was founded in 2018 by Geordie Rose, Suzanne Gildert, Ajay Agrawal, and Olivia Norton.[3] The company spun off from Kindred Systems Inc., a robotics firm that Rose and Gildert had co-founded in 2014. Kindred was notable for being one of the first companies to use reinforcement learning in a production robotics environment; it was later acquired by the Ocado Group in 2020.[4]
Geordie Rose brought significant technical credibility to Sanctuary. He co-founded D-Wave Systems in 1999, widely recognized as the first company to commercialize quantum computing, selling quantum computers to organizations including NASA and Google. Rose holds a PhD in theoretical physics from the University of British Columbia, specializing in quantum effects in materials.[5] His trajectory from quantum computing (D-Wave) to AI-driven robotics (Kindred) to general-purpose humanoid intelligence (Sanctuary) reflects a career focused on pushing the boundaries of computational systems.
Suzanne Gildert served as Sanctuary's co-founder and Chief Technology Officer, bringing experience from her work at both D-Wave and Kindred. In April 2024, Sanctuary announced Gildert's departure from the company.[6]
In November 2024, Sanctuary's board of directors removed Geordie Rose from the CEO position. James Wells, who had served as the company's Chief Commercial Officer for approximately five years, was appointed as interim CEO. The leadership transition was accompanied by staff reductions, and the company subsequently sought a $10 million convertible note offering to finance operations through fiscal 2025.[7][8]
Sanctuary AI's stated mission is to create "the world's first human-like intelligence in general-purpose robots." Rather than building specialized industrial automation, the company focuses on developing robots that can learn and perform a broad range of tasks in the same way a human worker would. This general-purpose approach contrasts with traditional industrial robots that are programmed for narrow, repetitive operations.
The company's strategy emphasizes dexterous manipulation over locomotion. While many competitors in the humanoid robot space prioritize bipedal walking and athletic capabilities, Sanctuary has consistently argued that hands capable of human-level manipulation are more commercially valuable than agile legs. As Geordie Rose explained in an interview, "Legs are nowhere near as important as hands" for the types of work tasks Phoenix is designed to perform.[9]
This thesis traces back to an observation that most economically valuable physical labor depends on fingers, not feet. Tasks such as sorting parts, fastening connectors, folding fabric, and packing boxes involve precise hand work within a small workspace, so a robot that stands or rolls in place while manipulating reliably can serve more industrial use cases than one that walks but cannot grip well.[20]
Phoenix has evolved through multiple hardware generations since Sanctuary began building robots. The company has iterated rapidly, releasing new generations approximately every six to twelve months.
| Generation | Year | Key developments |
|---|---|---|
| Generations 1 to 4 | 2018 to 2022 | Early research prototypes; internal development of core manipulation and AI systems |
| Generation 5 | 2022 to 2023 | First commercial pilot deployment at a Mark's retail store in Langley, BC; completed 110+ retail tasks |
| Generation 6 | May 2023 | Public unveiling of the "Phoenix" name; 170 cm, 70 kg humanoid form factor; introduction of Carbon AI; described as "first commercially available general-purpose robot" |
| Generation 7 | April 2024 | 50x improvement in autonomous task learning speed (tasks learned in under 24 hours instead of weeks); miniaturized hydraulics; improved hand durability and range of motion; enhanced visual acuity and tactile sensing |
| Generation 8 | December 2024 | Optimized for high-fidelity data capture; improved depth and vision cameras with wider field of view; enhanced telemetry; wheeled base replacing bipedal legs based on customer feedback; reduced bill of materials costs and simplified manufacturing |
Sanctuary's fifth-generation robot achieved a significant milestone in early 2023 when it was deployed at a Mark's retail store (owned by Canadian Tire Corporation) in Langley, British Columbia. During a week-long pilot, the robot successfully completed more than 110 distinct retail tasks, representing roughly 40% of all tasks performed in the store. These included picking and packing merchandise, cleaning, tagging, labeling, and folding items. The pilot allowed Canadian Tire to redirect human employees toward higher-value activities such as customer service.[10]
In May 2023, Sanctuary publicly introduced its sixth-generation robot under the name "Phoenix." Standing 170 cm (5 ft 7 in) tall and weighing 70 kg (155 lb), Phoenix established the humanoid form factor that subsequent generations would refine. This generation introduced the Carbon AI control system publicly and was accompanied by Sanctuary's description of the robot as the first commercially available general-purpose humanoid robot. Generation 6 featured hands with 20 degrees of freedom and a maximum payload capacity of 25 kg (55 lb).[11][12]
Phoenix was named to TIME magazine's list of Best Inventions of 2023, recognition that placed Sanctuary alongside more heavily funded humanoid programs and helped attract attention from prospective customers and investors. The TIME selection highlighted Phoenix's combination of human-like form factor and the cognitive flexibility provided by Carbon.[25]
Announced in April 2024, the seventh generation of Phoenix represented a major leap in autonomous learning capability. The time required for Phoenix to learn a new task dropped from several weeks to under 24 hours, a roughly 50-fold improvement. Engineers miniaturized the robot's hydraulics systems to reduce weight and power consumption while improving hand durability and range of motion. Visual acuity and tactile sensing were also upgraded.[13]
Geordie Rose described Generation 7 as "the most closely analogous to a person of any available" system, positioning it as "not only the cornerstone of general purpose AI robotics but a critical step on the path to artificial general intelligence."[14]
Released in December 2024, less than eight months after Generation 7, the eighth generation shifted focus toward high-fidelity data capture. Improvements included upgraded depth and vision cameras with greater field of view and resolution, enhanced telemetry systems, and improved audio and video capture. The generation was designed with manufacturing simplicity in mind, featuring reduced bill of materials costs and faster commissioning and assembly.[15]
Notably, Generation 8 adopted a wheeled base rather than bipedal legs. This design decision was informed by customer feedback indicating that bipedal legs were too fragile to support the strong torso needed for precise, safe work. For Sanctuary's target use cases (sorting parts, handling components, performing assembly tasks), manipulation capability takes priority over walking, and wheels offer a more reliable and practical mobility solution.[15]
| Category | Specification | Value |
|---|---|---|
| Physical | Height | 170 cm (5 ft 7 in) |
| Physical | Weight | 70 kg (155 lb) |
| Manipulation | Payload capacity | 25 kg (55 lb) |
| Manipulation | Degrees of freedom per hand | 21 |
| Manipulation | DOF per arm | 7 |
| Manipulation | Total upper-body DOF | ~63 (arms + hands) |
| Manipulation | Hand actuation | Miniaturized hydraulic valves |
| Manipulation | Fingers per hand | 5 |
| Manipulation | Tactile sensitivity | 5 millinewtons (mN) |
| Mobility | Top speed | ~3 mph (4.8 km/h) |
| Mobility | Locomotion (Gen 8) | Wheeled base |
| Mobility | Locomotion (Gen 6/7) | Bipedal |
| AI | Control system | Carbon |
| AI | LLM integration | Yes |
| AI | Natural language task instruction | Yes |
| Sensors | Vision | RGB cameras, depth cameras |
| Sensors | Tactile | 7-cell micro-barometer arrays per finger pad |
| Sensors | Force/torque | Yes |
| Sensors | IMU | Yes |
| Actuators | Type | Electric (body), hydraulic (hands) |
| Connectivity | Interfaces | WiFi, Ethernet |
Phoenix uses electric motors for the torso, arms, and locomotion base while reserving proprietary miniaturized hydraulic valves for the hands. Electric motors offer easy controllability for large joints but struggle to deliver the fingertip force and flow resolution needed for human-scale dexterity. Cable-driven hands used by several competitors are mechanically simpler but introduce friction and stretch that limit fine control as the cables wear. Sanctuary's choice of hydraulics for the hands trades added plumbing complexity for an order-of-magnitude improvement in power density and a faster bandwidth between commanded and applied force.[17]
Carbon is Sanctuary AI's proprietary cognitive architecture and serves as the "brain" controlling Phoenix. Sanctuary has described Carbon as the world's first commercial AI control system for a humanoid general-purpose robot.[11]
Carbon integrates multiple AI approaches into a unified cognitive architecture that mimics subsystems found in the human brain, including memory, sight, sound, and touch. The system combines:
This hybrid approach allows Carbon to translate natural language instructions into sequences of real-world physical actions. An operator or supervisor can describe a task in plain language, and Carbon will plan the steps, execute them, and adapt to unexpected situations. The system also features explainable and auditable planning, meaning its decision-making process can be inspected and understood by human operators.[11]
Sanctuary has described its broader research agenda as building toward Large Behavior Models (LBMs), an embodied analogue to the large language models that power modern chatbots. Where an LLM is trained on text to predict the next token, an LBM is trained on synchronized streams of sensor and motor data to predict the next physical action a competent worker would take. Sanctuary argues that the company's growing teleoperation dataset, captured across retail and factory pilots, gives Carbon a foundation of grounded behavioral examples that pure-vision or pure-language systems cannot match. Generation 8 was explicitly framed as a data-gathering platform optimized to feed this LBM training loop.[26]
A related concept from Sanctuary's research is the Eigengrasp: a small set of recurring hand postures that, when blended, can reconstruct most of the grip configurations humans use during manipulation. Identifying these basis patterns reduces the dimensionality of grasp planning and lets Carbon search a much smaller policy space than the raw 21-degree-of-freedom configuration would suggest.
Carbon fuses visual and haptic (touch) data to understand the physical properties of objects. Sanctuary's patented tactile sensor technology allows Carbon to distinguish between objects that look similar but have different physical properties. For example, the system can differentiate between picking up a fragile glass and a sturdy plastic cup of the same appearance, adjusting grip force accordingly.[16] This multimodal understanding of the physical world sets Carbon apart from AI systems that rely solely on vision.
Carbon uses photo-realistic and physics-realistic world simulations for robot training. These virtual environments allow Sanctuary to train policies in simulation and transfer them to the physical robot, reducing the need for costly and time-consuming real-world training sessions.[11]
Phoenix's hands are widely regarded as among the most advanced in any commercial humanoid robot program. Each hand features 21 degrees of freedom, approaching the approximately 27 degrees of freedom found in the human hand.[17]
Unlike most competitors that use cable-driven or electromechanical hand systems, Sanctuary employs proprietary miniaturized hydraulic valves for hand actuation. According to Sanctuary, this approach provides "an order of magnitude higher power density than cable and electromechanical-based systems," offering advantages in speed, strength, controllability, cycle life, impact resistance, and heat management.[17]
The durability of the hydraulic system has been validated through extensive testing. Sanctuary reported that its hydraulic valve actuators have been tested over 2 billion cycles without any signs of leakage or degradation, a significant milestone for long-term reliability in commercial applications.[17]
In February 2025, Sanctuary integrated a new generation of tactile sensors into Phoenix's finger pads. Each pad contains a 7-cell touch sensor array using micro-barometers, the same type of miniaturized pressure sensors found in smartphones, repurposed for robotic dexterity. Phoenix can detect forces as low as 5 millinewtons (mN). For comparison, human fingertip sensitivity is approximately 3 mN, which means Phoenix's sense of touch is within roughly 40% of human capability. This level of sensitivity far exceeds that of competitors relying solely on vision-based manipulation.[18]
The practical payoff of this sensitivity shows up in tasks where vision is partially or fully occluded. Sanctuary has highlighted blind picking (reaching into a bin where the target object is hidden), slip detection (recognizing the moment a grasped item begins to move and tightening the grip), and force-limited insertion (pressing a connector home without snapping it) as capabilities unlocked by the touch sensor upgrade. These behaviors are difficult for vision-only systems, which can only react after a slip becomes visible.[18]
Sanctuary demonstrated a significant in-hand manipulation milestone in late 2024, showing that Phoenix could simultaneously hold and reorient an object within its grasp. The robot successfully performed in-hand object reorientation even under extreme disturbance conditions, including an unexpected 500-gram load, making it one of the first commercial humanoid platforms to achieve this capability.[17]
In 2025, Sanctuary achieved successful zero-shot sim-to-real transfer for in-hand manipulation. This means an AI policy trained entirely in a simulated virtual environment performed successfully on the physical Phoenix hardware without any real-world fine-tuning. This result is particularly notable for contact-rich manipulation tasks, where many competing approaches require extensive real-world data collection or human-in-the-loop adjustment to bridge the gap between simulation and reality.[19]
The public demonstration used a lettered cube that Phoenix reoriented within a single hand, exposing each face in sequence. The cube task is a long-standing benchmark in robotic dexterity because it requires continuous fingertip regripping, the kind of behavior that usually degrades when policies move from simulation to hardware. Success without real-world fine-tuning suggested Sanctuary's simulated hand model captured the dynamics of the physical hand with high fidelity.[19]
Both the Carbon AI control system and the dexterous robotic hands are designed in a modular format. This enables potential future partnerships and integrations with both non-humanoid and humanoid general-purpose robots from other manufacturers.[17]
Sanctuary's approach to building autonomous robot capabilities is built on a distinctive teleoperation-to-autonomy training pipeline. Rather than attempting to program autonomous behaviors directly, the company uses human teleoperation as the foundation for generating training data that progressively enables autonomous operation.
The training pipeline operates in several stages:
Human motion capture: Sanctuary sensorizes human operators and records their movements while they perform useful work tasks. This captures the nuances of human manipulation, including grip adjustments, force modulation, and multi-step task sequencing.
Teleoperated robot deployment: The collected motion data informs the design and control of teleoperated Phoenix robots, which are deployed in real work environments. Skilled human operators control Phoenix remotely, performing productive tasks while the robot's sensors capture detailed data from every interaction.
Behavioral pattern identification: Sanctuary's AI systems analyze the teleoperation data to identify recurring tasks and subtasks, such as opening a door, grasping a handle, or placing an item on a shelf.
Autonomous task transition: Tasks or subtasks that are frequently repeated become candidates for autonomous execution. By automating these routine behaviors, the system significantly boosts the efficiency of the human pilot, who can focus on more complex or novel situations.
Sequence building and supervised autonomy: Individual autonomous tasks are chained into longer sequences, transitioning Phoenix from fully teleoperated mode to a supervised autonomy model where the robot executes multi-step work flows with human oversight.[20]
Every minute of Phoenix operation, whether teleoperated or autonomous, contributes to training data. Every sensor reading, motor position, and force measurement is recorded and transmitted for use in training Carbon AI models. Generation 8 of Phoenix was specifically optimized for this data capture role, with improved cameras, telemetry, and field of view designed to produce what Sanctuary calls "the highest quality, highest fidelity training data available."[15]
This approach creates a virtuous cycle: more deployments generate more data, which improves autonomous capabilities, which makes deployments more productive, which attracts more customers and more deployments.
In March 2025, Sanctuary disclosed a deepened collaboration with NVIDIA built around Isaac Lab, NVIDIA's open-source robot learning framework that runs on Isaac Sim. Isaac Lab lets Sanctuary spin up thousands of virtual Phoenix hands in parallel on GPU clusters, each exploring a slightly different manipulation strategy. Reinforcement learning policies that would take months to train on a single physical robot can be trained in days inside the simulator and then transferred to hardware. Sanctuary engineers reported that the simulated hands sometimes converge on motion patterns that go "beyond human capability," exploiting joint combinations a human teleoperator would not naturally produce.[27] This pairing of teleoperation (which captures realistic human strategies) with massively parallel reinforcement learning (which refines and extends those strategies in simulation) defines Sanctuary's current training stack.
In May 2024, Sanctuary announced a collaboration with Microsoft to accelerate AI development for general-purpose robots. Sanctuary uses Azure for the heavy lifting behind Carbon, including training runs for behavior models, inference workloads serving deployed Phoenix robots, and the networking and storage layers that connect remote teleoperation pilots to robots in customer facilities.[28]
Sanctuary's first commercial deployment took place in early 2023 at a Mark's retail store in Langley, British Columbia. Mark's is a subsidiary of Canadian Tire Corporation, one of Canada's largest retail groups. During the week-long pilot, the fifth-generation robot completed over 110 retail tasks including merchandise picking, packing, cleaning, tagging, labeling, and folding. This represented approximately 40% of all tasks required in the store, demonstrating that a general-purpose robot could meaningfully contribute in a retail environment.[10]
In April 2024, Sanctuary AI announced a strategic partnership with Magna International, a global automotive parts supplier headquartered in Aurora, Ontario. The multi-faceted collaboration includes:[21]
Magna has been an investor in Sanctuary AI since 2021, predating the formal partnership announcement. Magna operates automotive manufacturing facilities in multiple countries, including plants in Austria where it assembles vehicles for brands such as Mercedes-Benz, Jaguar, and BMW. In late 2025, Phoenix demonstrated its adaptability during trials at Magna's automotive plants, mastering new assembly tasks in under 24 hours through structured demonstrations of sorting and packing operations.[22]
In March 2024, Accenture made a strategic investment in Sanctuary AI through Accenture Ventures. As part of the investment, Sanctuary joined Accenture Ventures' Project Spotlight program, which connects emerging technology startups with Accenture's Global 2000 client base. Accenture identified significant potential for Phoenix in post and parcel handling, manufacturing, retail, and logistics warehousing operations.[23]
| Customer or partner | Sector | Year | Phoenix role |
|---|---|---|---|
| Mark's / Canadian Tire | Retail | 2023 | Picking, packing, cleaning, tagging, labeling, folding (110+ tasks) |
| Magna International | Automotive manufacturing | 2024 to present | Sorting small parts, handling wiring harnesses, sub-assembly, packing |
| Accenture Ventures | Consulting and integration | 2024 | Identifying use cases in post and parcel, logistics, retail |
| Microsoft | Cloud and AI infrastructure | 2024 | Azure-hosted training and inference for Carbon |
| NVIDIA | Simulation and GPU compute | 2025 | Isaac Lab-based parallel simulation training |
As of early 2026, Sanctuary AI has raised approximately $147 million across 12 funding rounds.[2]
| Round | Date | Amount | Key investors |
|---|---|---|---|
| Seed and early rounds | 2018 to 2021 | Undisclosed | Early-stage investors; Magna International |
| Series A | March 2022 | C$75.5M (US$58.5M) | Bell, Evok Innovations, Export Development Canada, Magna, SE Health, Verizon Ventures, Workday Ventures |
| Government grant (SIF) | November 2022 | C$30M | Government of Canada Strategic Innovation Fund |
| Strategic investment | March 2024 | Undisclosed | Accenture Ventures |
| Series A-V | July 2024 | US$11M | BDC Capital's Thrive Venture Fund, InBC Investment Corp. |
| Convertible note | January 2025 | US$2.5M | Undisclosed |
The C$30 million contribution from the Canadian government's Strategic Innovation Fund, announced in November 2022, was specifically directed at supporting the development of general-purpose robots with human-like intelligence for deployment in dangerous or labor-scarce work environments.[24]
Sanctuary has not published an official list price for Phoenix and has not opened the platform to general commercial sale outside of pilot relationships. Industry trackers and resellers have estimated unit prices ranging from roughly US$65,000 to US$250,000 for early integrated deployments, with the wide spread reflecting whether the figure includes Carbon licensing, teleoperation infrastructure, on-site support, and customization.[29] The company has publicly described its preferred go-to-market approach as "labor-as-a-service," where customers pay for delivered work output rather than purchasing robots outright, converting a large up-front capital expense into a recurring operating expense.[12]
Phoenix is built to operate near human workers rather than behind a safety cage, placing it within the category of collaborative robots, or cobots. Sanctuary states that Phoenix incorporates safety protocols aligned with ISO/TS 15066, the international technical specification that defines safety requirements for collaborative industrial robots, including power and force limits during contact with humans.[29] The wheeled base introduced in Generation 8 is also positioned as a safety improvement over earlier bipedal prototypes, since a tipped bipedal robot near a human worker is a more serious hazard than a wheeled platform that comes to rest on a stable footprint. The company has not yet published third-party certification documentation for all of the markets where it operates.
Phoenix operates in an increasingly crowded humanoid robot market. Several well-funded companies are pursuing general-purpose humanoid robots for commercial applications, each with distinct technical approaches and strategic focuses.
| Company | Robot | Key differentiator | Estimated funding |
|---|---|---|---|
| Sanctuary AI | Phoenix | 21-DOF hydraulic hands; Carbon AI; teleoperation-to-autonomy pipeline | ~$147M |
| Tesla | Optimus | Massive manufacturing scale; target price $20K to $30K; vertical integration | Internal (Tesla) |
| Figure AI | Figure 02 / Figure 03 | Partnership with OpenAI; advanced AI integration; bipedal locomotion | ~$750M+ |
| Boston Dynamics | Atlas (Electric) | Industry-leading mobility and agility; decades of locomotion research | Owned by Hyundai |
| Agility Robotics | Digit | Purpose-built for logistics; Amazon partnership; Robo-Fab manufacturing facility | ~$200M+ |
| Apptronik | Apollo | NASA heritage; modular design; Mercedes-Benz partnership | ~$80M+ |
| 1X Technologies | NEO | Consumer-focused; lightweight soft-body design; OpenAI-backed | ~$125M+ |
Sanctuary differentiates itself from competitors through several strategic choices:
Phoenix and Apptronik's Apollo share a similar height and weight class but pursue different technical bets. Apollo emphasizes electric force-controlled actuation and bipedal locomotion under a Mercedes-Benz partnership targeting general factory tasks. Phoenix emphasizes dexterous hand work and a teleoperation data pipeline, with its wheeled base trading bipedal walking for reliability. In hand performance, Phoenix's 21-degree-of-freedom hydraulic hands with 5-millinewton tactile sensitivity exceed the published specifications of Apollo's electric grippers; in mobility, Apollo sits closer to peers like Atlas and Digit.[29]
However, Sanctuary faces challenges relative to larger competitors. Tesla's Optimus benefits from massive manufacturing scale and could achieve dramatically lower per-unit costs. Figure AI has secured substantially more venture funding and a partnership with OpenAI for AI capabilities. Boston Dynamics brings decades of robotics research and the brand recognition of its Atlas platform. Sanctuary's smaller scale (approximately $147 million raised versus billions available to some competitors) means it must be highly strategic in its technology investments and market positioning.[2]
Phoenix is designed as a general-purpose platform applicable across multiple industries:
Sanctuary has described its business model as "labor-as-a-service," suggesting that customers would pay for robot work output rather than purchasing robots outright, although the company has also discussed direct sales.[12]
Phoenix has attracted sustained attention from technology press, industry analysts, and standards bodies since its 2023 unveiling. TIME magazine's Best Inventions of 2023 selection brought Phoenix to a general audience and emphasized the cognitive flexibility provided by Carbon.[25] IEEE Spectrum profiled the platform in 2023 with a focus on its teleoperation-to-autonomy roadmap, treating it as a leading example of the layered-autonomy strategy now common across the industry.[20] In February 2025, Morgan Stanley Research placed Sanctuary among the top three humanoid programs by published patent filings, a measure that captured the company's depth in hand and manipulation IP even as its venture funding lagged better-capitalized rivals.[30]
Reception has not been uniformly positive. Reviewers and trade press have noted that Phoenix remains in pilot deployment rather than at-scale commercial sale, that the leadership transitions of 2024 raised governance questions, and that Sanctuary's funding base is small compared with Tesla, Figure, and Boston Dynamics.[29] These structural concerns coexist with widespread industry agreement that Phoenix's hands and tactile sensing belong in the top tier of any commercial humanoid program.