Underactuation

The defining characteristic of an underactuated system is that the number of independent control inputs (actuators) is strictly less than the number of configurable degrees of freedom (DOF). For example, a robotic hand with 15 joints but only 6 motors is underactuated. This mismatch means the system cannot command every joint to an arbitrary position simultaneously; joint motions are coupled through mechanical linkages, tendons, or gears. Control must therefore reason about the passive dynamics of the unactuated joints.

Underactuation often introduces inherent mechanical compliance. When an underactuated gripper contacts an object, passive joints can conform to the object's shape without explicit sensing or control. This makes such systems highly adaptable to uncertain object geometries and contact locations. The trade-off is a loss of precise, independent control over each contact point. This principle is central to adaptive grippers and soft robotics, where compliance is a desired feature for robust, unstructured interaction.

The motion of underactuated systems is frequently governed by non-holonomic constraints—restrictions on velocity that are not integrable into positional constraints. A classic example is a car: it can move forward/backward and steer, but cannot move directly sideways. This means the system's path to a goal is not arbitrary; it must follow feasible trajectories that respect these constraints. In manipulation, a finger with coupled joints exhibits similar path-dependent behavior, requiring sophisticated nonlinear control and motion planning.

In underactuated systems, the dynamics of actuated and unactuated joints are strongly coupled through inertial and Coriolis forces. An action on one joint creates reaction forces that affect all others. This coupling can be exploited to induce natural motions like swinging or brachiating. Control strategies such as partial feedback linearization or energy shaping are used to orchestrate these complex interactions. The famous Acrobot and Pendubot are canonical underactuated systems studied for their dynamic richness.

A primary engineering motivation for underactuation is the significant reduction in hardware cost, weight, and control complexity. Fewer motors mean lower power requirements, reduced wiring, and less onboard electronics. This is critical for applications like space robotics, prosthetics, and mobile manipulators where weight and power are at a premium. The control challenge is shifted from hardware to software, requiring more advanced algorithms to manage the reduced actuation authority effectively.

Underactuation introduces fundamental control challenges. A system may not be fully controllable in all configurations; some states may be unreachable. Control laws must account for internal dynamics—the behavior of unactuated states. Techniques like backstepping, sliding mode control, and model predictive control (MPC) are commonly employed. Furthermore, underactuated systems often cannot hold arbitrary static poses, as gravity may dominate the unactuated joints, leading to a focus on dynamic stabilization rather than static positioning.

A Series Elastic Actuator is a key enabling technology for underactuated hands. It places a compliant spring element in series with the motor. This design provides inherent force control and shock absorption, allowing an underactuated finger to safely conform to objects and distribute contact forces without complex sensing. SEAs are fundamental for building robust, adaptive, and inherently safe underactuated systems.

Impedance Control is a fundamental control strategy for underactuated systems. Instead of dictating precise position, it regulates the dynamic relationship between force and motion, making the robot behave like a programmable mass-spring-damper system. This is ideal for underactuated hands because it allows fingers to naturally yield and adapt upon contact, enabling stable grasps across a variety of objects without explicit force closure calculations for every joint.

Force Closure is the geometric and force condition that determines if a grasp can resist arbitrary external wrenches (forces and torques). For an underactuated hand, achieving force closure is a primary design goal. The hand's mechanical coupling and passive compliance must be engineered so that when it closes on an object, the contact forces automatically configure to satisfy force closure, ensuring a secure grip despite having fewer actuators than degrees of freedom.

Non-Prehensile Manipulation involves moving objects without a enclosing grasp, using actions like pushing, pivoting, or tumbling. Underactuated hands often excel at these behaviors. Their passive mechanics and compliance allow them to perform dextrous in-hand manipulation—such as rolling a pen between fingers—by leveraging gravity, controlled slip, and environmental contacts. This expands their functional capability far beyond simple pinching.

Tactile Servoing is a closed-loop control method that uses real-time tactile sensor feedback to guide manipulation. For an underactuated hand, this feedback is critical for compensating for its reduced direct control. By sensing contact location, pressure, and slip, a controller can adjust the hand's few actuators to maintain a stable grasp, follow a contour, or perform fine repositioning, effectively closing the loop that the mechanical design leaves open.

The Grasp Wrench Space is the set of all possible wrenches a robotic hand can apply on an object through its contact points. Analyzing this space is crucial for evaluating underactuated hand designs. A robust underactuated hand is designed so that its mechanically coupled joint motions generate a wrench space that fully surrounds the origin, guaranteeing force closure for a wide range of object shapes and sizes within its enveloping grasp.