VP of Robotic Foundation Model
About the job
- Proven experience leading or managing high-performing machine learning, robotics artificial intelligence, or multimodal foundation model teams.
- A strong track record of taking advanced artificial intelligence or robotics systems from research or prototype stage into reliable real-world operation.
- Experience owning team execution, technical direction, prioritization, and stakeholder alignment for complex engineering programs.
- Demonstrated ability to lead in environments requiring both deep technical contribution and strong management.
- Deep expertise in designing, training, and evaluating large-scale multimodal models, such as vision-language, vision-language-action, or related transformer-based systems.
- Strong understanding of modern training paradigms, model scaling, fine-tuning, representation learning, and inference optimization.
- Experience integrating artificial intelligence or machine learning systems with physical robots under real-world operational constraints.
- Strong understanding of robotics software stacks, robot sensing, action representation, and the practical realities of deploying learned systems on hardware.
- Familiarity with robotics middleware such as ROS1/2 and embedded or edge artificial intelligence deployment platforms such as Jetson.
- Strong experience building data pipelines and training systems for large, complex multimodal datasets including images, video, text, robot trajectories, and sensor logs.
- Familiarity with distributed training frameworks and production machine learning infrastructure.
- Solid understanding of how high-level model decisions interact with low-level robot execution, control, safety, and system boundaries.
- Strong engineering judgment in balancing research ambition with deployment practicality.
- Strong communication skills in English.
- Lead the strategy, organization, and technical execution of the robotics foundation model team.
- Build and manage the team that turns advancing multimodal artificial intelligence technology into robust, production-grade capabilities for real-world robots.
- Define the roadmap for robotic foundation model efforts, guide the end-to-end pipeline from teleoperation data to on-robot deployment, and ensure the team delivers reliable, scalable models useful in real operational environments.
- Work across artificial intelligence, robotics, teleoperation, controls, hardware, and product teams to translate business and product goals into technical strategy and a high-performing engineering organization.
- Build, lead, and grow the robotics foundation model team, including hiring, mentoring, performance management, and team capability development.
- Define the team roadmap, set priorities, allocate resources, and drive execution toward business-critical milestones.
- Partner with Robotics, Teleoperation, Hardware, Controls, Infrastructure, and Product teams to align technical work with product and operational goals.
- Establish execution standards for planning, experiment review, quality, reproducibility, and deployment readiness.
- Own the technical direction for robotic foundation models integrating vision, language, robot state, and action.
- Guide architecture decisions, training strategy, and system design to balance model capability, reliability, and deployment practicality.
- Ensure model outputs integrate cleanly and safely with real robot control and autonomy systems.
- Define data strategy for teleoperation and robot-operation data, including collection, curation, annotation, and dataset quality.
- Oversee pipelines that transform raw multimodal robot data into training-ready datasets and evaluation assets.
- Drive continuous learning approaches so models improve reliably as new deployment data is collected.
- Lead deployment of trained models onto embedded and edge platforms such as Jetson-class systems.
- Define evaluation frameworks, key performance indicators, and review mechanisms for model quality, autonomy performance, safety, and operational robustness.
- Ensure failures observed in testing or the field are systematically analyzed and translated into model, data, or system improvements.
- Act as the company's technical leader for robotics foundation model development, influencing adjacent teams and executive decision-making.
- Represent the team in discussions with research partners, technology vendors, and external collaborators.
- Stay current with advances in multimodal artificial intelligence, robotics learning, and large-scale model systems, and apply relevant insights to the team roadmap.
- Japanese proficiency is a plus.
About the company
TELEXISTENCE designs and operates AI-powered robots under a robotics-as-a-service model to automate labor tasks in retail and logistics. Its TX SCARA robot restocks beverages using an AI system called Gordon with NVIDIA Jetson edge AI and can handle up to about 1,000 bottles or cans per day; a remote human operator can take control via telepresence if the AI struggles. The company combines AI, remote operation, and ongoing service deployments to scale automation for clients, supported by funding and partnerships with firms like FamilyMart, SoftBank, Foxconn, and Airbus Ventures. Its goal is to expand mass production and roll out versatile robots across retail and logistics, especially into North America, to reduce labor shortages and increase efficiency.
Company Size
51-200
Company Stage
Series B
Total Funding
$197.5M
Headquarters
Tokyo, Japan
Founded
2017
Get referred to TELEXISTENCE
See people who can refer or advise you
Simplify's Take
What believers are saying
- Ricoh invested July 16, 2026, expanding retail and logistics commercialization channels.
- Seven-Eleven Japan targeted Astra deployments for 2029, validating multi-chain demand.
- June 10, 2026 Fellowship results showcased single-policy humanoid autonomy on real stores.
What critics are saying
- Telexistence still depends on slow, task-specific grocery restocking; scaling to profit remains unproven.
- Astra's 2029 store rollout leaves years of execution risk before revenue diversification.
- If teleoperation costs stay high, robot economics fail and TX becomes a data-collection lab.
What makes TELEXISTENCE unique
- TX combines RaaS, teleoperation, and AI, proven in 300 FamilyMart stores.
- Its 2026 Motion Data Factory monetizes real-robot data, not just hardware.
- NVIDIA Cosmos Coalition membership and DreamZero work strengthen its physical-AI stack.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Health Insurance
Flexible Work Hours
Growth & Insights and Company News
6 month growth
1%1 year growth
3%2 year growth
-1%Ricoh has invested in Telexistence through its RICOH Innovation Fund, forming a capital alliance to develop AI-powered robotic automation for retail and logistics. Telexistence operates TX GHOST and TX SCARA robots that automate beverage restocking in convenience stores via a Robot-as-a-Service subscription model, already deployed by major Japanese retailers. The partnership combines Ricoh's digital services and AI expertise with Telexistence's robotics platform and remote operation technology. Telexistence uses a hybrid approach merging AI with human-assisted remote operation, enabling robots to function in dynamic retail environments whilst collecting data to improve AI models. Ricoh and Ricoh Japan will jointly explore research, development, and commercialisation opportunities in Physical AI, targeting intelligent automation solutions that address labour shortages and rising workforce costs.
Telexistence announces Physical AI Fellowship 2026 results. June 10, 2026 Unveiling Results of VLA-Based Single-Policy Autonomous Control and Joint Implementation with NVIDIA World Model "DreamZero" Telexistence Inc. (TX) today unveiled the results of its participation in the Physical AI Fellowship 2026, a global program powered by Amazon Web Services(AWS), NVIDIA and MassRobotics that helps high-potential robotics stratups build, refine, and scale physical AI solutions. TX is releasing a VLA-based autonomous manipulation demo on its humanoid robot, alongside the results of joint implementation with NVIDIA's World Model "DreamZero." Under the theme "From Real Humanoid Data to Physical AI," TX is publicly demonstrating for the first time an end-to-end pipeline that feeds large-scale, real-world teleoperation data directly into foundation model training - closing the loop from physical experience to Physical AI. 1. A single-policy model autonomously bags convenience store items. The core achievement of this demo is not a sequence of chained subtasks - it is a single VLA model (single policy) processing perception to action end-to-end. One model. One pass. No handoffs between sub-modules. The demo footage shows TX's humanoid robot at a Japanese convenience store checkout counter, grasping multiple items - PET bottles, onigiri, snacks - with both arms and placing them into a shopping bag. The "Autonomous" label onscreen confirms zero teleoperation intervention throughout. The single-policy approach is a deliberate architectural bet for scaling Physical AI. TX is going all-in on this direction. 2. Joint Implementation with NVIDIA "DreamZero" - bridging real data and model learning. TX is also releasing results from its joint implementation work with NVIDIA's World Model "DreamZero." Through the Physical AI Fellowship, TX has been working alongside the NVIDIA and AWS teams to implement DreamZero with a focus on real-world applicability. DreamZero is a World Model that predicts future states of a robot and its surrounding scene. In addition to predicting behavior in simulated environments, TX has demonstrated offline prediction from real humanoid teleoperation data - validating that data collected in the physical world can be fed directly into foundation model training. A World Model with grounded understanding of real-world dynamics has the potential to enable Physical AI systems to handle novel situations more robustly. TX continues this work in close collaboration with NVIDIA and AWS teams, positioning it as central to its robotics foundation model development. 3. Where Tx Inc is - and what Tx Inc is solving next. TX is releasing this demo not as a finished product, but as an honest snapshot of where the work stands today. The current technical challenges are clear - and they define TX's next development agenda. Autonomous operation has not yet reached practical deployment speed. The demo footage includes both 1x and 8x playback - a transparent acknowledgment that speed is the next major hurdle. Translating slow, human-demonstrated teaching data into fast, robot-optimized motion is the top priority on the technical roadmap. Motion Smoothness Current motion output still exhibits jerkiness. TX is actively working on how to best convert VLA model outputs into final robot control commands, as well as optimizing action representation in the VLA model itself and its training dataset. Generalization A shift of just a few centimeters in object placement can cause a drop in task success. TX believes this is fundamentally a data volume and diversity problem - solvable through scale. The open question of exactly how much data is needed is itself an important development challenge. TX's end-to-end ownership of hardware design, manufacturing, and real-world robot operation gives it a unique edge in pursuing the quantity and diversity of data needed to close this gap. Joint implementation with NVIDIA on DreamZero is also a key lever for improving data efficiency. 4. Why join TX now. Every challenge described above is an unsolved problem that the world's top institutions and companies are actively working on. TX is tackling them head-on with a unique set of advantages: * Real-world scale: TX already operates robots in live commercial convenience store environments - accumulating real-world data that academic labs simply cannot access. * Full-stack ownership: Hardware, data collection, foundation models, and control all under one roof - meaning implementation results go to real robots immediately. * Infrastructure: Large-scale training on state-of-the-art GPU clusters including AWS, simulation infrastructure via IsaacSim in collaboration with the NVIDIA team, and direct access to AWS Generative AI Innovation Center engineers. * Tight feedback loop: World model development through DreamZero and real-robot deployment run in the same team. Tx Inc is looking for engineers who want to work somewhere that's building the answers - not just applying them. Open positions. * Head of Robotics System Engineering, Humanoid Team * VP of Robotic Foundation Model * Logistics Robotics Hardware Engineer - Next Gen Products * Product Manager - Retail System * Engineering Manager - Store Applications "This demo is a starting point, not a finish line. Getting a single-policy VLA working on real hardware is a milestone - but the walls of speed, smoothness, and generalization are right in front of us. I want to work on these questions alongside people who are ready to face them head-on." - Genki Sano, Co-Founder & CTO, Telexistence About Physical AI Fellowship. The Physical AI Fellowship is a global program co-led by AWS, NVIDIA, and MassRobotics, designed to help high-potential robotics startups mature promising prototypes into real-world deployments. The 2026 cohort selected only 9 companies worldwide. TX is the first Japanese company to be selected. What fellows receive: * $200,000 in AWS credits * Embedded support from AWS Generative AI Innovation Center scientists and engineers * Access to NVIDIA Isaac and Cosmos frameworks (via NVIDIA Inception program) * Ecosystem support and facility access through MassRobotics * Demo opportunities at major events including AWS re:Invent 2026 About Telexistence. Telexistence Inc. (TX) is an AI robotics company on a mission to "grasp every single object in the world with our own hands." TX develops a full-stack platform integrating robotics foundation models, World Models, advanced manipulation AI, and purpose-built hardware - deploying intelligent robots across retail, logistics, and manufacturing. Press inquiries: Email: [email protected]
C-store teams with Telexistence to develop humanoid "Astra".
Tokyo, Japan, September 29, 2025 - Telexistence Inc. (TX) and Seven-Eleven Japan Co., Ltd. (SEJ) have entered into a partnership to jointly develop, validate, and introduce humanoid robots powered by generative AI.
Telexistence Inc. (Headquarters: Tokyo, CEO: Jin Tomioka) has been awarded the prestigious Prime Minister's Award at the government-led Japan Startup Awards 2025.