Legged robots have fascinated engineers since the earliest days of robotics. The appeal is obvious: legs allow traversal of terrain that wheels cannot handle — stairs, rubble, forest floors, and rocky slopes. But the engineering challenge is enormous. Legged locomotion requires solving balance, foot placement, energy efficiency, and contact dynamics simultaneously, in real time.
The ASIMO Era (2000–2018)
Honda's ASIMO, first demonstrated in 2000, represented the state of the art in humanoid locomotion for over a decade. Standing 130 cm tall and weighing 54 kg, ASIMO could walk, run at 9 km/h, climb stairs, and even carry a tray. Its control system used zero moment point (ZMP) theory to keep the center of pressure within the support polygon defined by its feet.
ASIMO proved that humanoid walking was achievable, but it also revealed the limitations of the ZMP approach. The robot required flat, known surfaces and could not handle disturbances well. A gentle push could cause it to fall. Honda retired ASIMO in 2022, but its legacy shaped every humanoid that followed.
Boston Dynamics and Dynamic Locomotion
Boston Dynamics took a fundamentally different approach. Rather than maintaining static stability, their robots — BigDog, Atlas, and Spot — embrace dynamic movement. Atlas can run, jump, do backflips, and recover from pushes using whole-body control that treats the robot as a single dynamic system rather than a collection of independent joints.
Spot, released commercially in 2020, became the first legged robot to achieve widespread industrial adoption. It is used in construction site monitoring, power plant inspection, mining, and public safety. Spot's success is as much about practical engineering — hot-swappable batteries, modular payload rails, a robust API — as it is about locomotion research.
The Quadruped Boom
Spot's commercial success triggered a wave of quadruped development. Unitree's Go1 and Go2 brought similar capabilities at a fraction of the cost, making legged robots accessible to researchers and hobbyists. Ghost Robotics' Vision 60 targeted defense and security applications. ANYbotics' ANYmal focused on industrial inspection in hazardous environments like offshore platforms.
What unites these platforms is the convergence of three technologies: high-torque actuators (quasi-direct-drive motors that provide both speed and compliance), inertial measurement units for balance, and onboard computing powerful enough to run model predictive control or reinforcement learning policies in real time.
Humanoids Return
By 2025, humanoid robots re-entered the spotlight. Tesla's Optimus, Figure's 02, Agility Robotics' Digit, and 1X's NEO are all targeting human-scale bipedal robots for logistics, manufacturing, and domestic assistance. Unlike ASIMO, these new humanoids leverage learning-based control, compliant actuators, and whole-body planning to achieve more natural, robust movement.
The economic argument is compelling: a humanoid that can operate in spaces designed for people — warehouses, homes, offices — doesn't require infrastructure modifications. Whether the technology will mature fast enough to deliver on these promises remains an open question, but the investment flowing into the space suggests the robotics community is betting heavily on legs.
