Join Lumos

Responsibilities: 1. Develop and optimize world-model algorithms for robotic perception and decision-making. 2. Work on multimodal data fusion, spatiotemporal representation learning, and reinforcement-learning simulation. Requirements: 1. Strong command of PyTorch or JAX, with deep understanding of world models and generative models. 2. Familiarity with robotics and reinforcement learning. 3. Publications at top conferences or major competition awards are preferred. Our Beijing office is newly established. Compensation and title are open-ended for outstanding candidates, and we welcome conversations with exceptional algorithm talent.

Core Responsibilities: 1. Multimodal model development Deep understanding of Transformer architectures, with hands-on capability in training and tuning VLA, VLM, and LLM systems. Familiarity with leading approaches such as RT-2, OpenVLA, OpenEMMA, and EMMA is expected. Master multimodal alignment methods to enable end-to-end generation from vision and language inputs to robot action sequences, including task planning and motion control. 2. Optimization and deployment Proficient with PyTorch, DeepSpeed, and other large-model frameworks, with experience in distributed training across multiple machines and GPUs. Optimize inference efficiency on embedded platforms through lightweight deployment and CUDA acceleration, supporting real-time control for physical robots or autonomous driving systems. 3. Data and training strategy Leverage imitation learning (IL) and reinforcement learning (RL) to improve model generalization in physical environments. Requirements: Master's degree or above in computer science, artificial intelligence, robotics, or related fields. Practical deployment experience in industrial robotics, humanoid robotics, or autonomous driving is preferred. Strong grasp of frontier technologies and emerging research directions. At Lumos, we are building a long-term ecosystem for general embodied intelligence and believe robots can drive the next wave of intelligent transformation. Powered by our FastUMI data ecosystem and robot platform, we are continuously iterating a multimodal data stack that spans human-centric capture, ego-view data, dexterous hand manipulation, and force-tactile information, together with general foundation models. This forms a full loop from data collection to model training, real-world deployment, and data flywheel feedback. Our general embodied AI system is already deployed in broad industrial scenarios, allowing algorithm research to be validated and iterated directly in real environments. You will work on pre-training and post-training for foundation models, VLA systems, world models, dexterous manipulation, imitation learning, reinforcement learning, and AI infrastructure research. We take a patient, long-term approach to core technical foundations and are committed to building a sustainable general embodied AI system that keeps getting better over time. The company is well capitalized and can support long-horizon research and front-line innovation.

Role Overview: Lead architecture design and algorithm R&D for the motion-control system of next-generation intelligent robots, especially wheeled dual-arm platforms. Solve core technical challenges in high-precision force control, full-stack motion planning, and whole-body control (WBC) under complex operating conditions, and help move products from prototype to mass production. Core Responsibilities: Full-stack motion-control framework Lead the design of a complete motion-control software stack covering high-level business orchestration, path and trajectory planning, obstacle-avoidance optimization, and low-level execution. Advanced control algorithms Develop hierarchical multi-task controllers based on HQP, QP, and CBF to unify end-effector tracking, redundancy optimization, and obstacle avoidance. Build time-optimal trajectory planners that support dynamic replanning under joint velocity, acceleration, and torque constraints. High-precision force-control system Build force-control pipelines based on joint torque feedback or six-axis force sensors. Develop high-performance disturbance observers (DOB) to suppress friction and transmission interference. Implement hybrid force-position control, impedance control, and adaptive force-control algorithms to improve stability and precision in contact-rich tasks. Modeling and identification Perform high-precision dynamics modeling, parameter identification, and singularity handling, and use systematic analysis to optimize control stiffness and bandwidth. Embodiment and teleoperation integration Develop master-slave teleoperation systems and use tools such as trajectory filters to interpolate and smooth low-frequency embodied commands. Simulation and tooling Build simulation systems with MuJoCo, PyBullet, and similar frameworks. Develop automated tuning software and controller debugging and analysis tools. Requirements: Master's degree or above in control engineering, robotics, aerospace, or related disciplines. Core Skills: Expert-level proficiency in C/C++ and Python, with familiarity in real-time control frameworks such as ROS, ROS2, and OROCOS/RTT. Strong robotics fundamentals, including spatial transformation, forward and inverse kinematics, redundant analytic IK, dynamics, and force-control theory. Experience with optimization libraries such as Eigen and Pinocchio, and with physics engines. Project Experience: Hands-on experience developing collaborative robots, wheeled mobile manipulators, or parallel force-control robots is preferred. Candidates with successful mass-production programs or strong publication records at top conferences such as Humanoids are highly valued. General Capability: Ability to independently build integrated software and hardware test platforms and translate frontier methods such as DMP and CBF into practical engineering systems.

Core Responsibilities: 1. Mechanical design and optimization Own the mechanical design of joint modules, including integration of transmission systems such as harmonic drives and RV reducers, as well as motors and encoders, ensuring performance targets in load, precision, and stability are met. Use finite element analysis tools such as ANSYS and ABAQUS to evaluate strength, stiffness, and modal behavior of key components and continuously optimize the structure. 2. Control algorithms and system development Participate in control-system development, including controller selection, C/C++ programming, and algorithms such as PID and torque control, to improve motion precision and responsiveness. Work closely with electrical engineers on interface design and wiring plans to ensure coordinated electromechanical performance. 3. Testing and production support Define test plans and validate key module performance metrics such as repeatability and load capacity. Solve technical problems during production and support prototyping, commissioning, and manufacturability optimization for mass production. Requirements: 1. Education Bachelor's degree or above in mechanical engineering, mechatronics, automation, or related fields. A master's degree is preferred for higher-complexity roles. Solid knowledge of mechanical design, material mechanics, automatic control theory, motors, and drive systems is required. 2. Experience At least 3 years of experience in robot joint modules or reducer R&D, with successful product-delivery experience preferred. Familiarity with the working principles of robot joint modules and their critical subsystems. 3. Skills Proficient with design tools such as SolidWorks and AutoCAD, with the ability to produce manufacturing-ready engineering drawings. Experience with finite element analysis and dynamic modeling tools such as MATLAB and Simulink. Familiarity with PLCs, microcontrollers, servo drives, and related control hardware.

This role will take full ownership of the company's overseas sales strategy and execution, with a primary focus on Southeast Asia, Europe, and North America. We are especially interested in candidates with deep experience in overseas education-tech markets or robotics markets, and those from the embodied AI industry are strongly preferred. The role will lead strategy development, sales-system building, key-channel expansion, and team execution to achieve revenue targets and scale product deployment and brand presence globally. Key Responsibilities: 1. Strategy and execution Study regional trends, policy environments, customer needs, and competitive landscapes across Southeast Asia, Europe, and North America. Build medium- and long-term overseas sales strategy, annual revenue goals, and regional expansion plans, break them down to team and individual targets, and ensure disciplined execution. 2. Major accounts and complex deal ownership Personally manage top overseas customers, including major educational institutions, robotics integrators, manufacturing enterprises, and education groups. Lead negotiations, contracting, and delivery for strategic deals while coordinating the full project lifecycle and driving scalable replication in target sub-segments. 3. Brand and compliance management Collaborate with marketing to grow Lumos brand awareness and visibility overseas. Stay current on regional regulations, industry standards, and trade rules to ensure compliant sales operations and proactively manage commercial, IP, and regulatory risks. Requirements: 1. At least 5 years of overseas sales management experience. Candidates must be familiar with either the overseas education market, including K12, universities, and research procurement workflows, or overseas robotics markets. Embodied AI background is preferred. You should have end-to-end regional operating experience in Southeast Asia, Europe, or North America, with successful examples in channel expansion, key-account development, and target delivery. 2. Strong market understanding of sales models, customer characteristics, industry pain points, compliance expectations, and product-market fit within the target regions.

Key Responsibilities: 1. Government relationship development and expansion Build on existing relationships with local government stakeholders such as science and technology bureaus, industry and information bureaus, investment promotion offices, and data bureaus. Maintain senior decision-maker relationships while proactively expanding target city-level government and enterprise accounts, and align local industrial-planning priorities with Lumos data-factory construction opportunities. 2. Government communication and demand interpretation Understand and operate fluently within government communication frameworks. Accurately interpret policy direction and decision logic related to digital economy, compute infrastructure, data elements, and investment attraction. Independently produce formal presentations, feasibility studies, and cooperation proposals that meet government standards and support reporting and review processes. 3. Full-cycle leadership of data-factory projects Lead the full lifecycle of jointly built government data-collection-field projects, including demand discovery, solution design, business negotiation, project filing, budget approval, tendering, contracting, delivery, landing, and payment follow-up, ensuring complete closed-loop execution. 4. Commercial negotiation and risk control Independently lead negotiations for major government-level projects. Be highly familiar with public-sector collaboration rules, bidding processes, and compliance requirements. Manage commercial clauses, cooperation models, and investment-sharing structures while identifying and mitigating policy, commercial, and compliance risk. 5. Internal and external coordination Coordinate across R&D, product, delivery, and pre-sales teams to respond quickly to government needs and provide tailored data-collection solutions. Align with government departments and ecosystem partners to keep projects moving efficiently. 6. Ecosystem development and opportunity generation Track trends across compute infrastructure, data collection, and digital economy industries. Analyze competitors and market direction, identify new opportunities through government resources, and replicate the data-factory cooperation model across more cities to deliver business targets. Requirements: 1. At least 5 years of major-account sales experience in government and enterprise settings, preferably in computing centers, intelligent computing centers, data centers, or adjacent sectors. 2. Rich and actionable local-government resources, with deep relationships at city level or above in departments such as MIIT, science and technology bureaus, and data bureaus. Familiarity with government organization, decision-making, procurement, and successful public-sector project delivery is required. 3. Strong command of government-facing communication logic, with the ability to independently deliver high-level proposals and lead business negotiations. Outstanding resource integration, negotiation, resilience, and execution ability are essential for winning large government-led projects end to end.