Modern Robotics, Part 2B — Frames, Lie Groups, and the Jacobian Bridge

In the first half of this post, we introduced twists, screw axes, and the exponential map. The main message was that a rigid-body motion is generated by an infinitesimal motion generator and that finite motion can be obtained by exponentiating that generator.

This second half focuses on the remaining subtleties that matter most in practice:

• how the same motion is represented in different frames,
• why the Lie group / Lie algebra viewpoint is so natural,
• and why the Jacobian is the natural next object after the screw axis.

1. The same motion, different coordinates

A rigid-body motion can be expressed in different coordinate frames. This is one of the most common sources of confusion in robotics.

If a twist is expressed in the spatial frame, we write it as

$$\xi_s.$$

If it is expressed in the body frame, we write

$$\xi_b.$$

These are not different motions; they are the same physical motion described from different reference frames.

The relationship between them is given by the adjoint transformation:

$$\xi_s = Ad_T(\xi_b),$$

or equivalently,

$$\xi_b = Ad_{T^{-1}}(\xi_s).$$

This is an important reminder: the motion is invariant, but the coordinates are not.

In other words, a twist is not “just a vector.” It is a vector in a particular representation of the rigid-body motion algebra, and this representation depends on the frame.

2. Why frame choice matters

Frame choice matters because the same physical motion may look different when represented in coordinates attached to the world or to the moving body.

For example, suppose the end effector is moving relative to the robot base. The world-frame representation is useful when the goal is to describe motion in the environment. The body-frame representation is useful when the goal is to describe motion relative to the tool itself.

This distinction shows up directly in Jacobian design:

• the spatial Jacobian uses quantities expressed in the spatial frame,
• the body Jacobian uses quantities expressed in the body frame.

The two formulations are mathematically equivalent, but they lead to different matrices and different computational conventions.

This is why we often say that the geometry is the same and only the representation changes.

3. Lie groups and Lie algebras in plain language

The rigid-body transform group is

$$SE(3).$$

Its corresponding infinitesimal motion algebra is

$$se(3).$$

The group is the set of all finite rigid transformations. The algebra is the set of all infinitesimal motions represented by twists.

The exponential map connects these two objects:

$$T = e^{\hat{\xi}}.$$

This is the mathematical reason why the move from instantaneous motion to finite motion works so cleanly.

The Lie algebra is not just an abstract object; it is exactly the set of motion generators. A twist is the algebra element, and the exponential map turns that algebra element into a finite transform in the group.

This is also why robot kinematics has such a beautiful structure: the geometry of finite motion is encoded by the algebra of infinitesimal motion.

4. The matrix logarithm as the inverse idea

The matrix logarithm gives the reverse direction:

$$\hat{\xi} = \log(T).$$

This is useful when we want to take a known rigid transformation and recover the corresponding infinitesimal motion generator.

In some robotics problems, we may have a measured or commanded pose and want to understand the local twist that generated it. The logarithm gives exactly that local description.

This is particularly important in estimation and control. We often do not need the full rigid-body transformation history; we only need the local velocity generator that best explains the current change.

5. Why the exponential map is the right language for joints

A robot joint does not produce arbitrary motion. It produces motion along a specific screw axis. Therefore, each joint contributes a generator of the form

$$\hat{\mathcal{S}}_i \theta_i.$$

The total end-effector motion is then composed as

$$T_{sb}(\theta) = e^{\hat{\mathcal{S}}_1 \theta_1} e^{\hat{\mathcal{S}}_2 \theta_2} \cdots e^{\hat{\mathcal{S}}_n \theta_n} T_{sb}(0).$$

This is the Product of Exponentials formula, which is often considered one of the most elegant formulations in robotics.

The key point is not just the algebraic formula. The point is conceptual:

• each joint contributes a geometric motion generator,
• the robot’s full motion is the product of these generators,
• the structure is inherited from the geometry of rigid-body motion itself.

This is why Modern Robotics feels so coherent: the kinematic model is written in the same language as the geometry of motion.

6. From screw axes to Jacobians

Once we have screw axes, the next object naturally emerges: the Jacobian.

The Jacobian is simply the matrix whose columns are the joint screw axes, expressed in the required frame:

$$J(\theta) = \begin{bmatrix} \mathcal{S}_1 & \mathcal{S}_2 & \cdots & \mathcal{S}_n \end{bmatrix}.$$

Then the end-effector twist is

$$\nu = J(\theta)\dot{\theta}.$$

This is not a separate idea added on top of screw theory. It is the direct consequence of assembling the motion generators of each joint into a single linear map.

The Jacobian is exactly the object that tells us how joint rates produce task-space motion.

That is why it is so central to robotics: it is the first step from geometric motion description to robot control.

7. Geometric intuition behind the Jacobian

The Jacobian is often treated as a numerical matrix, but geometrically it is richer than that.

Its columns represent how each joint moves the end effector in instantaneous task space. If the columns are aligned or nearly dependent, the manipulator cannot move independently in all directions. That is the geometry of singularity.

This is the intuition behind the rank condition:

$$\operatorname{rank}(J) < n.$$

If the rank is deficient, some joint motions do not create independent task-space motion.

Therefore, the Jacobian connects three levels at once:

• geometric motion directions,
• instantaneous task-space velocity,
• and the local conditioning of the manipulator.

This is one reason why the screw-theory viewpoint is so powerful: it turns a matrix into a physical explanation.

8. A compact conceptual summary

The full flow of this series can now be summarized in a compact form:

1. rigid-body motion is represented by a transform $T \in SE(3)$,
2. instantaneous motion is represented by a twist $\xi \in se(3)$,
3. a joint generates motion along a screw axis $\mathcal{S}$,
4. finite motion comes from the exponential map $e^{\hat{\mathcal{S}}\theta}$,
5. the Jacobian assembles all joint contributions into a single map $\nu = J\dot{\theta}$,
6. this linearized map is the basis for control, optimization, and planning.

This is the conceptual backbone of Modern Robotics.

9. Why this matters beyond theory

It is easy to think of these ideas as mathematical elegance alone. But in reality, everything from trajectory tracking to inverse kinematics to force control depends on this structure.

The reason is simple:

• the robot’s motion is rigid-body motion,
• instantaneous motion is a twist,
• finite motion is an exponential,
• and task-space velocity is a Jacobian-weighted combination of joint rates.

This is not an optional viewpoint. It is the language in which modern robot kinematics is written.

10. Final takeaway

The most important lesson from this split is that the screw-axis formulation is not an isolated chapter. It is the foundation for the entire robot kinematic model.

Twists describe instantaneous motion, screw axes describe the geometry of joint motion, and the exponential map connects those infinitesimal motions to finite transformations.

Once this is understood, the Jacobian is no longer mysterious. It is simply the matrix that assembles the motion generators of all joints into a single velocity map.

That is the bridge from geometry to control.