Legged Robot

A robot that moves on legs and performs terrain adaptation and balance control; quadruped, bipedal, and humanoid forms are expanding into industrial, service, and disaster-response settings.

Legged Robot

Overview

A legged robot is a robot that moves using legs instead of wheels. Leg structures provide high adaptability on unstructured terrain such as stairs, slopes, gravel, and debris, and balance control and locomotion planning are core technologies. Recently, advances in reinforcement learning and simulation-based learning, high-torque actuators, and lightweight sensor fusion are rapidly driving the practical deployment of quadrupedal and humanoid robots.

Main Content

Definition and Classification

Legged robots are divided by number of legs into bipedal, quadrupedal, and multi-legged robots. Bipedal locomotion is compatible with human environments but has high difficulty in balance control, while quadrupedal locomotion has excellent stability and travel performance and was commercialized first in industrial settings. Humanoids combine bipedal locomotion with upper-body manipulation capabilities, aiming for general-purpose tasks in human workspaces.

Core Technologies

1) Actuators: There are electric, hydraulic, and artificial-muscle methods; recently, high-torque-density electric actuators and series elastic actuation are attracting attention. 2) Balance control: IMU, joint encoders, and force/torque sensors are fused to estimate the center of mass and ground reaction forces in real time. 3) Locomotion planning: This requires foothold selection, stride length and speed adjustment, and terrain-awareness-based foot placement planning. 4) Perception: LiDAR, stereo cameras, and depth sensors recognize surrounding terrain and obstacles, and SLAM estimates position. 5) Learning-based control: A method of learning reinforcement learning and imitation learning in simulation and then transferring to real robots (sim-to-real) is becoming common.

Representative Cases

Boston Dynamics' Spot is a representative commercial model of a quadrupedal robot and is used for industrial inspection and patrol. Atlas is a symbolic platform for humanoid research and has recently transitioned to an electric-drive version. China's Unitree is expanding the market with low-cost quadrupedal and humanoid robots, and ANYbotics, Hyundai Motor Company, Tesla, Figure, and Agility Robotics are also developing legged robots for industry, logistics, and service.

Applications

Representative applications include industrial equipment inspection, construction site monitoring, logistics warehouse transport, disaster rescue, agricultural patrol, medical and rehabilitation assistance, and education and research platforms. In particular, legged robots are highly valuable in hazardous areas that are difficult for humans to access or in repetitive patrol tasks.

Challenges

Battery energy efficiency, high-cost components, safety certification, response to severe weather and unknown terrain, securing large-scale data, and norms for human-robot collaboration remain challenges to be solved. In addition, for humanoids, hands and tactile technologies for general-purpose manipulation and long-term operational stability are important.

Latest Trends

In 2024-2025, competition for commercialization of humanoid robots began in earnest. Tesla Optimus, Figure, and Agility Digit were deployed in factory and logistics site pilots, and robot learning combined with AI foundation models is spreading. Large-scale simulation learning using NVIDIA Isaac Sim·Lab, reinforcement-learning-based whole-body control, and research on robot agents that follow natural-language commands are active. Price competition has intensified with Chinese companies' launches of low-cost quadrupedal and humanoid robots, and commercial deployment of industrial quadrupedal robots is increasing in regular inspection and safety patrol fields. On the regulatory side, discussions on robot safety standards and liability are also underway.

Related Topics

  • [[Humanoid robot]]
  • [[Quadrupedal robot]]
  • [[Reinforcement learning]]
  • [[Robotics]]
  • [[Boston Dynamics]]