Configuration Space (C-Space)

The dimensionality of C-Space directly corresponds to the robot's Degrees of Freedom (DOF). A 6-axis industrial manipulator has a 6-dimensional C-Space. Each axis represents an independent joint parameter, such as a revolute angle or prismatic displacement. The topology of this space is not always Euclidean; a revolute joint creates a circular dimension, making the space a generalized cylinder or torus. This high-dimensionality is the core computational challenge, as the volume of the search space grows exponentially with DOF, a phenomenon known as the curse of dimensionality.

Physical obstacles in the robot's workspace are mapped into C-Space as Configuration Space Obstacles (C-Obstacles). A C-Obstacle is the set of all robot configurations that cause a collision with a physical object. This mapping accounts for the robot's entire geometry, not just its end-effector. The remaining space is C-Free, the set of all collision-free configurations. Path planning then becomes the problem of finding a continuous curve entirely within C-Free, connecting the start and goal configurations.

C-Space captures not just position but also motion constraints. A holonomic system can move instantaneously in any direction within its C-Space; a robotic arm is a classic example. A nonholonomic system, like a car or differential-drive mobile robot, has velocity-level constraints that cannot be integrated into position constraints. This means the robot cannot slide sideways, and its C-Space must be augmented with a tangent bundle to represent feasible directions of motion, making path planning significantly more complex.

The topology of C-Space can introduce singularities, configurations where the robot loses one or more degrees of freedom. At a singularity, the mapping from joint velocities to end-effector velocities becomes degenerate, requiring infinite joint rates to achieve a finite Cartesian velocity. These are internal obstacles in C-Space that must be avoided. Furthermore, C-Space may not be simply connected; obstacles can create 'holes' or 'tunnels', requiring planners to explore narrow passages that are notoriously difficult for sampling-based algorithms to navigate.

C-Space is rarely constructed explicitly due to its high dimensionality. Instead, it is probed implicitly. A collision checker acts as a binary oracle: given a configuration, it returns 'collision' or 'free'. Sampling-based planners like RRT and PRM exploit this by randomly sampling configurations and testing them, building a graph of free configurations without ever computing the full boundary of C-Obstacles. This implicit representation is what makes planning in high-dimensional spaces computationally tractable.

A valid path in C-Free is not enough; we seek an optimal or high-quality path. This requires defining a distance metric in C-Space. A naive Euclidean metric in joint space is often misleading, as a small joint change for a base axis can cause a large end-effector displacement. Weighted metrics or Riemannian metrics that account for link lengths and inertia are used. The cost function can also incorporate time, energy, or smoothness, transforming the geometric path into a trajectory optimization problem.