北京机科未来科技有限公司

【Job Responsibilities】 1. Algorithm Design and Development • Responsible for designing and implementing reinforcement learning algorithms for optimizing robot grasping tasks. • Design task objectives and reward mechanisms for robot grasping tasks, and deploy reinforcement learning algorithms for large-scale parallel training. 2. Simulation and Testing • Build a large-scale parallel simulation environment for robots, used for algorithm testing and verification. • Test and optimize reinforcement learning algorithms in a simulation environment to ensure algorithm stability and effectiveness. 3. Technological Research and Innovation • Track the latest research progress in the field of embodied intelligence and reinforcement learning, and reproduce the algorithm framework of the latest research results. • Carry out cutting-edge research on reinforcement learning related to robot motion control and manipulation, and explore new algorithms and methods. 4. Project Collaboration and Implementation • Collaborate with robot hardware engineers, software engineers, and other team members to deploy reinforcement learning algorithms onto actual robot platforms. • Promote the application of reinforcement learning algorithms in embodied intelligence large models and other directions, accelerating the transformation of algorithms from theory to practice. 5. Data Processing and Analysis • Collect and process data related to robot grasping tasks for algorithm training and optimization. • Analyze experimental data, evaluate algorithm performance, and propose improvement plans.

1. Education Background • Master's degree or above in Computer Science, Automation, Robotics Control, or related fields. 2. Professional Skills • Familiar with mainstream reinforcement learning algorithms, such as PPO, SAC, DDPG, etc. • Possesses a solid foundation in deep learning, reinforcement learning, robot kinematics and dynamics, and automatic control principles. • Proficient in Python or C++ programming languages, and familiar with Linux/ROS operating systems. • Familiar with at least one mainstream robot simulation environment or framework, such as IsaacGym/Sim, Gazebo, MuJoCo, etc. 3. Project Experience • Experience with reinforcement learning projects related to robot control. • Preference given to candidates with experience in robot grasping projects. 4. Academic Achievements • Preference given to candidates with publications in robotics conferences such as RSS, CoRL, ICRA, IROS, or AI conferences such as CVPR, NeurIPS, ICLR. 5. Other Abilities • Excellent communication, collaboration, problem-solving, quick learning, self-motivation, and innovation skills. • Excellent English reading and writing skills.

1. Develop motion control algorithms based on reinforcement learning/model predictive control (MPC) to achieve stable movement on complex terrain. 2. Design a multi-rigid-body dynamics simulation system to optimize gait planning and balance control for humanoid robots/robotic dogs. 3. Develop an AI-driven real-time motion decision system to enhance adaptability to dynamic environments. 4. Build a digital twin testing platform to complete transfer learning from algorithm simulation to physical robots. 5. Participate in hardware-algorithm joint debugging to optimize sensor fusion and actuator control strategies.

1. Master's degree or above in Robotics, Automation, Computer Science or related fields. 2. Proficient in C++/Python, familiar with ROS/ROS2 development framework. 3. Practical experience in at least one of the following areas: • Whole-body motion control algorithm development for humanoid robots • CPG/MPC gait control implementation for quadruped robots • Trajectory planning and hybrid force/position control for robotic arms 4. Familiar with physics simulation tools such as MuJoCo/Isaac Gym. 5. Preference given to candidates with experience in reinforcement learning (PPO/SAC) and optimal control projects. 6. Additional points: • Participation in benchmark projects such as Boston Dynamics Atlas, Unitree Unitree • Publication of papers in top conferences such as RSS, ICRA, IROS • Familiarity with key technologies such as humanoid robot state estimation and compliant control

1. Develop a general-purpose embodied algorithm and apply it to humanoid robot scenarios, with the ability to generalize objects, tasks, and scenes; 2. Research multi-modal embodied large models with visual, tactile, and language perception and decision-making capabilities to control robots to complete physical interactions in open worlds; 3. Technological Research and Innovation • Track the latest research progress in embodied intelligence and reinforcement learning, and reproduce the algorithm frameworks of the latest research results. • Conduct cutting-edge research on reinforcement learning related to robot motion control and manipulation, exploring new algorithms and methods. 4. Project Collaboration and Implementation • Collaborate with robot hardware engineers, software engineers, and other team members to deploy reinforcement learning algorithms onto real robot platforms. • Promote the application of reinforcement learning algorithms in embodied intelligence large models, accelerating the transformation of algorithms from theory to practice. 5. Data Processing and Analysis • Collect and process data related to robot grasping tasks for algorithm training and optimization. • Analyze experimental data, evaluate algorithm performance, and propose improvement plans.

1. Master's degree in Computer Science, Artificial Intelligence, Natural Language Processing, or related fields; doctoral degree preferred. Proficient in Python programming and familiar with deep learning frameworks such as TensorFlow and PyTorch; 2. Familiar with Transformer, BERT, GPT models, pre-training and post-training processes. Experience with large model training and optimization is required; 3. Preference given to candidates with research experience in algorithms such as ACT, MT-ACT, RT1, RT2, and Diffusion Policy; 4. Additional points: • Familiarity with reinforcement learning algorithms; • Publications in top conferences such as ICML, ICLR, CVPR, RSS, and ICRA.