B-Spline Trajectory

Control points are a set of points in the robot's configuration space (e.g., Cartesian space, joint space) that influence the shape of the B-Spline curve. The curve does not generally pass through these points but is "pulled" toward them. Their primary function is to provide local control; moving a single control point only affects a local segment of the curve, defined by the spline's degree. This property is crucial for real-time trajectory adjustments in dynamic environments.

• Example: In a 2D path for a mobile robot, control points might be (x, y) coordinates. A planner can adjust a point to avoid a newly detected obstacle without recomputing the entire path.

The knot vector is a non-decreasing sequence of real numbers that defines the parameter domain and the influence of control points. It partitions the parameter range into intervals where different polynomial basis functions are active. The spacing and multiplicity of knots determine the continuity (smoothness) of the spline at the junctions between polynomial segments.

• Uniform vs. Non-Uniform: A uniform knot vector has equally spaced knots, simplifying computation. A non-uniform knot vector allows for uneven spacing, providing greater flexibility to model complex shapes.
• Knot Multiplicity: Increasing the multiplicity of a knot (repeating its value) reduces the continuity at that parameter location. A multiplicity equal to the degree creates a discontinuity, which can be used to represent sharp corners or trajectory segments.

Basis functions, also called B-Splines, are the piecewise polynomial functions defined recursively by the Cox-de Boor algorithm, using the knot vector. Each control point is weighted by a corresponding basis function. The value of the trajectory at any parameter is the weighted sum of the control points, where the weights are given by these basis functions.

• Key Property: Convex Hull & Local Support: The basis functions are non-negative and partition unity (sum to 1), ensuring the curve lies within the convex hull of its active control points. Each basis function is non-zero only over a span of degree+1 knots, enforcing local control.

The degree p (or order k = p + 1) of a B-Spline defines the polynomial order of its segments and its inherent smoothness. A spline of degree p is C^(p-1) continuous (has p-1 continuous derivatives) at non-repeated knots, provided the knot vector is structured appropriately.

• Degree Selection Trade-off:
  • Low Degree (e.g., p=3, Cubic): Common choice. Provides C^2 continuity (continuous position, velocity, acceleration), which is often sufficient for smooth actuator commands. Computationally efficient.
  • High Degree (e.g., p=5): Provides higher-order continuity (e.g., continuous jerk or snap), which can be important for minimizing wear on mechanical systems or for specific dynamic constraints. Increases computational cost and reduces local control influence.

Parameterization refers to the mapping from the spline's parameter (often u in the range defined by the knot vector) to time. A uniform parameterization (where u advances linearly with time) is simple but may result in non-constant velocity along the path. For physical trajectories, a time-optimal or kinodynamic parameterization is critical.

• Time-Optimal Parameterization: This involves solving a separate optimization problem to assign time values to each knot, subject to actuator velocity and acceleration limits. The result is a B-Spline trajectory where the derivative with respect to time (the actual robot velocity) respects dynamic constraints.

The derivatives of a B-Spline (velocity, acceleration, jerk) are themselves B-Splines of a lower degree. Their control points can be computed analytically from the original control points and knot vector. This is a major advantage for motion planning.

• Efficient Constraint Enforcement: Kinodynamic constraints (e.g., max velocity v_max, max acceleration a_max) become simple bounds on the control points of the derivative splines. This allows for fast optimization where the trajectory is guaranteed to satisfy constraints everywhere, not just at sampled points.
• Example: For a cubic B-Spline position trajectory (degree 3), the velocity is a quadratic B-Spline (degree 2). Ensuring all velocity control points have magnitude less than v_max guarantees the entire trajectory respects the speed limit.

A defining advantage of B-Splines is local control. Modifying a single control point only affects the curve within a limited range defined by the basis functions and knot vector. This is critical for:

• Online trajectory repair: Adjusting a path segment to avoid a newly detected obstacle without recomputing the entire trajectory.
• Interactive design: Allowing engineers to fine-tune specific sections of a motion path for precision tasks.
• Real-time adaptation: Enabling dynamic systems to modify their planned motion locally in response to disturbances.

B-Splines provide mathematically guaranteed smoothness. A B-Spline of degree p has continuous derivatives up to order C^{p-1}. This is essential for physical systems where jerk (the derivative of acceleration) must be bounded.

• High-degree continuity: A cubic B-Spline (degree 3) guarantees continuous acceleration (C^2), producing smooth velocity profiles.
• Actuator and passenger comfort: Minimizes vibrations and sudden forces on mechanical joints and passengers in autonomous vehicles.
• Energy efficiency: Smooth trajectories typically require less peak torque and power from motors.

The convex hull property ensures the entire B-Spline curve lies within the convex hull of its control points. This enables efficient and conservative collision checking.

• Fast feasibility checks: By verifying that the convex hull of a subset of control points is collision-free, you can guarantee the corresponding curve segment is safe, without evaluating every point on the curve.
• Conservative corridor generation: Control points can be placed inside a verified safe corridor, ensuring the resulting trajectory remains within it.
• Integration with sampling-based planners: Paths from planners like RRT* can be represented as sequences of control points whose convex hulls are known to be in free space.

B-Splines provide a compact, differentiable parameterization ideal for trajectory optimization and Model Predictive Control (MPC).

• Low-dimensional parameterization: A long, complex curve is defined by a relatively small set of control points, reducing the number of decision variables in optimization.
• Analytic derivatives: Costs and constraints based on position, velocity, and acceleration can be computed efficiently as linear combinations of control points and basis function derivatives.
• Real-time solvers: This efficiency enables the use of Nonlinear Programming (NLP) solvers within the tight control loops of real-time robotic control systems.

In Computer Numerical Control (CNC) and robotic milling, B-Splines are the standard for defining complex toolpaths.

• Smooth surface finishing: The high continuity eliminates machine tool vibrations, producing superior surface quality on manufactured parts.
• Compact G-code: Modern CNC standards (e.g., NURBS, a generalization of B-Splines) allow dense toolpath contours to be transmitted with few control points, reducing program size.
• Precise cornering: Allows for blended cornering, where the tool smoothly rounds a sharp programmed corner at a controlled tolerance, reducing wear and cycle time.

B-Splines are extensively used for planning smooth, drivable paths for autonomous ground vehicles (AGVs) and unmanned aerial vehicles (UAVs).

• Dynamic feasibility: The smoothness directly relates to bounded curvature and derivatives, ensuring the trajectory respects the vehicle's nonholonomic constraints and actuator limits.
• Local replanning: The local control property allows the vehicle's local planner to quickly adjust the upcoming path segment when reacting to pedestrians or other dynamic obstacles.
• Formation control: For multi-robot coordination, B-Spline trajectories provide a shared parameterized path that individual agents can track with offset, maintaining fleet shape.

A minimum jerk trajectory is a path that minimizes the integral of the squared magnitude of the third derivative of position (jerk) over time. This optimization criterion produces exceptionally smooth, human-like motions that are comfortable for passengers and reduce mechanical stress on robotic joints.

• Mathematical Basis: Formulated as a boundary value problem, often solved via Pontryagin's maximum principle or direct collocation methods.
• Common Use: Widely used for robotic arm trajectories and autonomous vehicle lane-change maneuvers where passenger comfort is paramount.
• Comparison to B-Spline: While a B-Spline can represent a smooth path, a minimum jerk trajectory is defined by a specific optimality criterion. A B-Spline can be optimized to approximate a minimum-jerk path.

Trajectory optimization is the mathematical framework for finding a sequence of states and control inputs that minimizes a cost function (e.g., energy, time, jerk) while satisfying dynamic constraints and avoiding obstacles. It transforms a motion planning problem into a numerical optimization problem.

• Core Methods: Includes direct methods (shooting, collocation) and indirect methods.
• Role of B-Splines: B-Splines are frequently used as the parameterization for the trajectory in optimization. The control points become the decision variables optimized by the solver.
• Key Challenge: Balancing computational speed with solution feasibility and optimality, especially in real-time applications like Model Predictive Control (MPC).

Nonlinear Programming (NLP) is the process of solving an optimization problem where the objective function or some of the constraints are nonlinear. It forms the mathematical and computational backbone for solving trajectory optimization problems.

• Connection to Planning: When a B-Spline trajectory is optimized for smoothness or dynamic feasibility, the resulting problem is typically an NLP.
• Common Solvers: IPOPT, SNOPT, and CasADi are standard tools for solving the NLPs generated in robotics.
• Problem Structure: Constraints include collision avoidance (nonlinear inequality), dynamics equations (nonlinear equality), and actuator limits (bound constraints).

Model Predictive Control (MPC) is an advanced, real-time control method that repeatedly solves a finite-horizon trajectory optimization problem. It uses an internal dynamic model to predict future system behavior and applies only the first control input from the optimized sequence before re-planning.

• Real-Time Trajectory Generation: MPC inherently performs online trajectory optimization. B-Splines are often used within the MPC formulation to parameterize the predicted state or control trajectory.
• Key Advantage: Explicitly handles state and input constraints (e.g., velocity limits, obstacle boundaries).
• Computational Demand: Requires fast, reliable NLP solvers to complete optimization within the control loop's timeframe (often milliseconds).

Spline parameterization refers to the use of piecewise polynomial functions (splines) to represent continuous paths or trajectories with a finite set of parameters (e.g., control points, knots). It is a foundational technique for making continuous optimization problems tractable.

• Types of Splines: Includes B-Splines, Bezier curves, NURBS (Non-Uniform Rational B-Splines), and cubic splines.
• Why Parameterize?: Converts an infinite-dimensional search space (all possible continuous curves) into a finite-dimensional one, enabling numerical optimization.
• B-Spline Advantages: Offers local control (moving a control point affects only a local curve segment) and built-in smoothness (C^(k-2) continuity for order k), which simplifies constraint formulation.

Collision avoidance constraints are mathematical conditions enforced during motion planning or trajectory optimization to ensure a robot's path does not intersect with obstacles. For B-Spline trajectories, these constraints must be evaluated along the continuous curve.

• Common Representations: Obstacles are often represented as convex shapes, polyhedra, or implicit functions via Signed Distance Fields (SDF).
• Constraint Formulation: For a B-Spline path p(t), a simple constraint is SDF(p(t)) > safety_margin for all t. In practice, this infinite set of constraints is sampled or bounded.
• Efficient Evaluation: The convex hull property of B-Splines can be used to conservatively check for collisions by examining the convex hull of a subset of control points, speeding up planning.