Skip to main content

Chapter 34: Humanoid Manipulation

Design of Humanoid Hands and Arms​

Humanoid robots are designed with arms and hands that mimic the human musculoskeletal structure to enable dexterous interaction with the environment. The design prioritizes a balance between strength, reach, and precision.

Arm Structure​

Humanoid robot arms typically feature multiple degrees of freedom (DoF) to replicate human-like movements, including shoulder (pitch, roll, yaw), elbow (pitch), and wrist (pitch, yaw, roll) joints. These joints are often actuated by compact and powerful motors (e.g., servo or BLDC motors) embedded within the limb or transmitted via cables/tendons to keep distal segments lightweight.

Hand and Gripper Design​

The design of humanoid hands ranges from simple two-finger grippers to highly complex, anthropomorphic multi-fingered hands.

  • Anthropomorphic Hands: These hands closely resemble human hands, often featuring 3 to 5 fingers with multiple joints per finger. They are capable of various grasp types (power grasps, precision grasps) and complex in-hand manipulation. Materials like silicone or rubber are often used for fingertips to enhance friction and conformability.
  • Underactuated Hands: To reduce complexity and cost, some hands are underactuated, meaning fewer motors control more joints. This relies on the mechanical design of the hand to achieve adaptive grasping.

Grasping Strategies: Strong vs. Delicate​

Humanoid robots need versatile grasping capabilities to interact with a wide array of objects, from heavy tools to fragile items.

Power Grasps​

Power grasps involve the palm and all fingers enclosing an object, providing high stability and force. This is suitable for securely holding heavy or large objects (e.g., holding a hammer).

Precision Grasps​

Precision grasps use only the fingertips and thumb, allowing for fine control and manipulation of small or delicate objects (e.g., picking up a pen or a small electronic component).

Adaptive Grasping​

Modern grasping strategies often incorporate adaptive algorithms that use sensor feedback (e.g., tactile sensors, force sensors) to adjust grip force and finger posture dynamically, ensuring both secure and delicate handling.

Fine Motor Control​

Fine motor control is crucial for tasks requiring precision and dexterity. This is achieved through a combination of highly sensitive sensors and sophisticated control algorithms.

  • Tactile Feedback: Sensors embedded in the fingertips provide information about contact pressure and object texture, enabling the robot to refine its grip.
  • Force Control: Robots can regulate the force applied during grasping or manipulation, preventing damage to delicate objects or ensuring sufficient grip on slippery surfaces.
  • Vision-Guided Manipulation: High-resolution cameras provide visual feedback, allowing the robot to accurately locate, approach, and manipulate objects, especially in cluttered environments.

Tool Use​

Tool use is a significant capability that extends a robot's functional range. Humanoid robots can be trained to pick up and operate various human tools (e.g., screwdrivers, drills, pens). This requires:

  • Tool Recognition: Identifying the type of tool and its functional parts.
  • Grasp Planning: Determining the optimal grasp for operating the tool.
  • Task-Oriented Control: Executing sequences of movements that effectively utilize the tool to achieve a desired outcome.

In-Hand Manipulation​

In-hand manipulation refers to the ability to re-position or orient an object within the gripper without releasing and re-grasping it. This is a highly dexterous skill that humans perform effortlessly but is challenging for robots. Techniques include:

  • Finger Gaits: Coordinated movements of individual fingers to roll or slide an object.
  • Palm-Finger Synergies: Using both the palm and fingers to achieve complex object reorientation.

The advancement in humanoid manipulation allows robots to perform a broader range of complex tasks, bridging the gap between automated industrial actions and flexible human-like interaction.