6. Principles of Robotic Manipulation
The Art of Interaction
Locomotion gets a robot to the right place, but manipulation allows it to perform useful tasks once it's there. Manipulation is the art and science of physically interacting with objects: picking them up, moving them, and using them to achieve a goal. For a humanoid robot, this means controlling its arms and hands with precision and dexterity.
This chapter explores the core principles that govern how a robot controls its limbs to interact with the world.
The Arm: A Kinematic Chain
A robot's arm is a kinematic chain: a sequence of rigid links (the bones) connected by joints (like your elbow and shoulder). To control the arm, we must understand its geometry. This is the study of kinematics.
Forward Kinematics (FK): The "Easy" Question
- Question: If I know the angles of all my joints, where is my hand?
- Process: FK uses the robot's known link lengths and its measured joint angles (from encoders) to calculate the precise 3D position and orientation of the end-effector. It's a straightforward problem of applying a series of geometric transformations (translations and rotations).
Inverse Kinematics (IK): The "Hard" Question
- Question: If I want my hand to be at a specific target position, what should all my joint angles be?
- Process: This is a much harder problem. Unlike FK, which has one unique answer, IK can have:
- No solution: The target is outside the robot's reachable workspace.
- Multiple solutions: The robot can reach the same point with different poses (e.g., "elbow up" vs. "elbow down").
- Infinite solutions: If the robot is redundant (has more joints than necessary), there are infinite ways to reach a target.
IK is typically solved with iterative numerical optimizers that try to find the best joint configuration to minimize the distance between the hand and the target.
The Hand: End-Effectors
The "hand" of the robot is its end-effector.
- Simple Grippers: Most robots today use simple two-fingered parallel grippers. They are robust, simple to control, and effective for a wide range of objects.
- Dextrous Hands: Advanced, multi-fingered hands that mimic human dexterity are a major area of research. They offer incredible flexibility but are mechanically complex and extremely difficult to control.
A key distinction in grasping is the difference between a power grasp (wrapping the whole hand around an object for a secure hold) and a precision grasp (using the fingertips for fine-grained control).
From Position to Force: The Jacobian
So far, we've only talked about position. But manipulation is fundamentally about applying forces. How do we control the force exerted by the hand? The answer lies in a critical piece of mathematics: the Jacobian matrix (J).
The Jacobian is the bridge between the robot's joints and its end-effector.
- Velocity Mapping: It relates the velocity of the joints to the velocity of the hand.
end_effector_velocity = J * joint_velocities - Force Mapping: Through its transpose (
J^T), it also relates the force you want to apply with the hand to the torques you need at the joints.joint_torques = J^T * end_effector_force
The Jacobian allows us to switch from thinking purely in terms of position to thinking in terms of force.
The Importance of Feeling: Force Control and Compliance
Imagine trying to open a drawer with your eyes closed. You would push forward gently until you felt the handle, then pull. You are using force feedback. A robot must do the same.
A purely position-controlled robot is a poor manipulator. If it's commanded to move to a point inside a table, it will try to do so with maximum force, potentially breaking itself or the table. To perform contact tasks, a robot needs compliance.
Compliance is the ability to yield or "be soft" in response to external forces. It is essential for:
- Safety: A compliant robot is safer to be around.
- Contact Tasks: Inserting a key, wiping a surface, or turning a crank all require compliant control that balances position and force.
Hybrid Position/Force Control is a common strategy where the robot's control is split. For example, when sliding a block along a table, the controller might be in "position control" in the horizontal direction but in "force control" in the vertical direction, commanding the arm to always press down with a gentle, constant force.
Code Example: Forward Kinematics for a 2-Link Arm
Let's see how Forward Kinematics works with a simple 2D, two-link robot arm. Given the link lengths and joint angles, we can calculate the position of the hand.
- Python
import numpy as np
def forward_kinematics_2d(link_lengths, joint_angles):
"""
Calculates the (x, y) position of a 2D, 2-link robot arm's end-effector.
Args:
link_lengths (list or tuple): A list of two link lengths [l1, l2].
joint_angles (list or tuple): A list of two joint angles in radians [theta1, theta2].
theta1 is the angle of the first link from the base.
theta2 is the angle of the second link relative to the first.
Returns:
np.array: A 2D array [x, y] for the end-effector's position.
"""
l1, l2 = link_lengths
th1, th2 = joint_angles
# Position of the end of the first link (the "elbow")
x1 = l1 * np.cos(th1)
y1 = l1 * np.sin(th1)
# The angle of the second link is relative to the first, so we add the angles.
# The position of the end-effector is relative to the elbow.
x2 = x1 + l2 * np.cos(th1 + th2)
y2 = y1 + l2 * np.sin(th1 + th2)
return np.array([x2, y2])
# --- Simulation ---
link_lengths = [1.0, 0.8] # Link 1 is 1.0m, Link 2 is 0.8m
# Case 1: Both joints at 90 degrees (pointing straight up)
angles1 = [np.pi / 2, 0] # 90 degrees for th1, 0 for th2
pos1 = forward_kinematics_2d(link_lengths, angles1)
print(f"Angles {np.rad2deg(angles1)} deg -> End-effector position: {pos1.round(2)}")
# Case 2: First joint at 45 degrees, second bent inward by 90 degrees
angles2 = [np.pi / 4, -np.pi / 2] # 45 degrees for th1, -90 for th2
pos2 = forward_kinematics_2d(link_lengths, angles2)
print(f"Angles {np.rad2deg(angles2)} deg -> End-effector position: {pos2.round(2)}")