Robotics core · topic B2 · 40 hours
Forward/inverse kinematics & Jacobians
The native language of a robot arm. The Jacobian also decides «which directions cannot push», which makes it the basis of singularity analysis and force control.
What to learn
- Building a 6-axis model with DH parameters and writing the forward-kinematics transform chain
- Two ways to solve inverse kinematics: closed-form (geometric / algebraic) versus numerical (iterative), and when each applies
- The Jacobian's two readings: joint velocity → tip velocity, and tip force → joint torques
- Singular configurations: why joint speeds blow up as det(J) approaches 0
- Redundant DOFs and the null space: what else the arm can do while the tip stands still
The lesson · plain words
Forward kinematics: given every joint angle, compute where the tip is. All multiplication, no surprises. Inverse kinematics runs the other way: given where the hand must go, back out the joint angles. This step has no unique answer — it may have none at all, or it may land on a singularity that sends a motor overspeed in an instant.
This page covers both directions, plus the matrix that connects «how fast the joints turn» to «how fast the tip moves»: the Jacobian. It is the shared entry point of force control, optimization, and calibration.
Forward: multiply down the chain
Every joint contributes one 4×4: revolute joints give rotations, prismatic joints translations, with fixed link offsets in between. Multiply from base to tip down the chain: the last column is the tip position, the upper-left 3×3 is the tip orientation.
That is exactly what a teach pendant displays. It is also why changing one link length shifts the whole tip position — you edited one link of the chain.
Inverse: two solutions, out of reach, singular
Two links already give two solutions: elbow-up and elbow-down. Reach too far and there is no solution — the reach boundary is a hard wall. Fully stretched, the Jacobian determinant approaches 0: the tip wants to move 1 cm and the joints must turn tens of degrees. That is a singularity.
Drag the point in the figure and produce all three cases: the solution jumping between branches, no solution outside the circle, and readings blowing up at full extension.
Interactive figure · drag it and see
The Jacobian: the exchange rate between joint speed and tip speed
The Jacobian is a pose-dependent matrix: tip velocity = J × joint velocity. It translates «how fast the motors turn» into «how fast the hand moves» — and translates forces back the other way.
A determinant near zero is a singularity: the exchange rate collapses, tip speed in some direction approaches zero while the joint speeds it would take approach infinity. Every «the motor screamed at one particular position» story is this one row of numbers.
Interactive figure · drag it and see
How inverse kinematics is really solved in industry
Closed-form: classic 6-axis geometries (spherical wrist) admit closed formulas — fast, enumerable over all 8 solutions, the industrial default. Numerical: damped least-squares or QP — general, but iterating every millisecond. Optimization: joint limits, collision avoidance, and torque caps all folded into one constraint solve.
The engineering default without closed form is «numerical + damping»: the damping coefficient is the tip error you accept in exchange for joint speeds that never explode. Better to miss by a millimeter than to spin a motor to 3000 rpm.
Three questions — do you really get it?
1. How many inverse-kinematics solutions does a 2-link planar arm generally have?
- AExactly one
- BTwo (elbow-up / elbow-down)
- CInfinitely many
Show answer
B. A planar 2-link arm has two mirror solutions. A 6-axis industrial arm has up to 8; the controller picks the one nearest the current pose.
2. As the Jacobian determinant approaches 0, what is the most dangerous effect?
- AFor the same tip speed, joint speeds approach infinity
- BThe tip position computes wrong
- CThe orientation representation gimbal-locks
Show answer
A. A vanishing determinant means some tip directions become unmovable, and the reverse solve amplifies joint speeds toward infinity. On hardware: instant overspeed and a protective stop.
3. The target sits outside the reach circle. What does a typical position controller do?
- AErrors out and stops
- BShortens its own links
- CRails against the limit — dangerous under force control
Show answer
C. Most position controllers keep emitting saturated commands, the arm pressing at its limit. In position mode it buzzes; in force mode it can crush the workpiece.
Keep going deeper
Links without a marker stay on this site; the rest point to the Chinese edition until its translation batch lands.
How it fails in the field
- No singularity guard on the inverse solution: joints snap near a singular pose and the field crew blames the drives
- Solving position but not orientation: approach angles come out random, grasp success never improves, and nobody finds the cause
You have learned enough when
The Chinese edition carries the full topic page for Forward/inverse kinematics & Jacobians, including the complete outline, depth requirements per career track, and further derivations: robotguide.dev/k/B2/.