Quadruped Robots

Definitional Boundaries & Taxonomy

Quadrupedal locomotion represents the pinnacle of underactuated, high-dimensional non-linear control systems. Unlike wheeled platforms confined to contiguous 2D manifolds, legged robots exploit discrete footholds, enabling navigation across highly unstructured, discontinuous 3D terrains. Taxonomically, quadruped robots are classified into two primary morphological architectures: mammalian and reptilian. Mammalian configurations feature legs articulated directly beneath the torso, optimizing sagittal plane movements and gravitational load-bearing efficiency. Reptilian configurations utilize sprawled postures, lowering the Center of Mass (CoM) for extreme stability on hyper-inclined planes but suffering from higher continuous joint torque requirements.

Furthermore, quadrupeds are divided by their stabilization regimes into statically stable and dynamically stable platforms. Static stability relies on maintaining the CoM projection strictly within the support polygon defined by at least three grounded feet (e.g., creep gait). Dynamic stability, conversely, allows the robot to momentarily violate static equilibrium, utilizing inertial forces to balance during aerial phases or minimal-support phases (e.g., trot, pace, bound, gallop).

Historical Evolution

The genesis of quadrupedal robotics is rooted in the 1960s with the General Electric Walking Truck, a master-slave hydraulic behemoth requiring immense human cognitive bandwidth to operate. The fundamental paradigm shift occurred in the 1980s via Marc Raibert's Leg Laboratory at MIT. Raibert decoupled the highly complex legged control problem into three orthogonal sub-tasks: vertical hopping (energy state), forward speed control, and posture control.

The 2000s witnessed the inception of Boston Dynamics' BigDog, utilizing an onboard internal combustion engine driving hydraulic pumps. While historically significant for proving off-road utility, its acoustic signature and hydraulic inefficiencies forced a pivot. The 2010s marked the golden era of electric actuation. The MIT Cheetah and ANYbotics' ANYmal introduced proprioceptive quasi-direct drive (QDD) and Series Elastic Actuators (SEA), drastically improving energy regeneration and mechanical transparency. In the 2020s, the commoditization of high-density BLDC motors and advanced compute paradigms has led to an explosion of commercial systems (e.g., Unitree Go2/B2, Ghost Robotics Vision 60), shifting the core challenge entirely from hardware viability to software robustness.