Robotics, Vision & Control 3e: for Python (Springer, 2023) was finished in 2022. Its prime dependencies — RTB, MVTB, bdsim, spatialmath-python — have all evolved since. Where a notebook's code has to change to keep working, that's a divergence from what's printed on the page, so it's recorded here.
Scope: only entries for changes a reader actually has to make to get a
notebook cell running as intended. Bugs that were fixed upstream (in RTB,
MVTB, bdsim, or spatialmath-python) with no notebook-side change
don't belong here — the printed book code still matches the notebook in
those cases, once the dependency is updated. Keep this list lean; it's not
a running log of every toolbox bug found while testing, only the ones that
actually changed a line the reader would see. (For the fuller
investigation log — including issues that turned out not to need a
notebook change — see claude-notes/notebook-testing-issues.md.)
The same applies to a renamed API that keeps a working deprecated alias:
if the printed code still runs against the released packages, perhaps
with a DeprecationWarning, it isn't an erratum (the notebooks were still
updated). Those are listed under Not listed
at the end. The reader-facing version of this file, with book page numbers
and links to the fixing and causing PRs, is the
Errata wiki page.
One ## entry per changed cell/line group, in reading order. Section
numbers come from the nearest preceding section-heading markdown cell in
the notebook itself (e.g. # 2.3.1.1 3D Rotation Matrix).
## §<section number> — <short description of what changed and why>
**Notebook:** `chapN.ipynb`
**Reason:** <one line — which dependency changed and how, or what broke
and why>
**As printed:**
```python
<code exactly as it appears in the book>
```
**Current toolbox syntax:**
```python
<code needed today>
```Notebook: chap7.ipynb
Reason: RTB deprecated and removed models.list(); the replacement
is models.catalog(), and the type-filter kwarg was renamed from type
to mtype (type shadows the Python builtin, which was the reason for
the rename).
As printed:
models.list(type="ETS")
models.list(type="URDF")
models.list(type="DH")Current toolbox syntax:
models.catalog(mtype="ETS")
models.catalog(mtype="URDF")
models.catalog(mtype="DH")Notebook: chap13.ipynb
Reason: Not a toolbox API change -- a transcription error. rho=1.4e-6
is in metres, so f must be too; f=3045 (3045 metres) makes fx = f/rho
absurdly large (~2.18e9 pixels), which makes every cv2.solvePnP call
inside scene.fiducial() silently fail. The figure-generation script that
produced the book's own printed figure (figures/code/chapter13/ fig13_31.py) uses the correct value, f=4.25e-3.
As printed:
camera = CentralCamera(f=3045, imagesize=scene.shape,
pp=(2016, 1512), rho=1.4e-6);Current toolbox syntax:
camera = CentralCamera(f=4.25e-3, imagesize=scene.shape,
pp=(2016, 1512), rho=1.4e-6);Notebook: chap13.ipynb
Reason: Not a toolbox API change -- a plain (non-raw) Python string
containing \p and \t isn't a recognized escape sequence, so this raises
SyntaxWarning: invalid escape sequence '\p' (and would for \t too, but
\t happens to be a legitimate escape -- tab -- so only \phi warns; both
should be raw strings for correctness).
As printed:
spherical.disp(axes=("$\phi$", "$\theta$"));Current toolbox syntax:
spherical.disp(axes=(r"$\phi$", r"$\theta$"));Notebook: chap14.ipynb
Reason: MVTB's Image class moved to an immutability model (commit
9afa287a7, "Image immutability"): paste() used to modify the receiver
in place (when called with its old default copy=False) and this is what
the book's own figure-generation script relied on
(figures/code/chapter14/fig14_49.py, which predates this change). It now
always returns a new Image and leaves the original untouched, matching
every other Image method's current functional style -- the copy=
parameter is retained on the signature but ignored. Calls that don't
capture the return value are now silent no-ops: composite never
accumulates the pasted tiles, so every downstream cell in this section
(SIFT on an all-zero canvas, homography estimation, the second paste())
fails or produces nonsense.
As printed:
composite.paste(images[0], (0, 0));
composite.disp();
...
composite.paste(tile, topleft, method="blend");
composite.disp();Current toolbox syntax:
composite = composite.paste(images[0], (0, 0));
composite.disp();
...
composite = composite.paste(tile, topleft, method="blend");
composite.disp();Notebook: chap12.ipynb
Reason: NumPy 2.0 made np.linalg.eig() always return complex-dtype
arrays, even when every eigenvalue's imaginary part is exactly zero
(NumPy 1.x returned real dtype in that case). J here is a real
symmetric matrix (built from image moments), so its eigenvalues are
always real -- np.linalg.eigh() is both the numerically correct choice
for a symmetric matrix and sidesteps this NumPy 2.0 change, since it
guarantees real output.
As printed:
lmbda, x = np.linalg.eig(J)Current toolbox syntax:
lmbda, x = np.linalg.eigh(J)Notebook: app.ipynb
Reason: Same NumPy 2.0 change as §12.1.3.3: np.linalg.eig() now
always returns complex-dtype arrays, even when every eigenvalue's
imaginary part is exactly zero (NumPy 1.x returned real dtype in that
case). E here is a real symmetric matrix (the ellipse's defining
matrix), so its eigenvalues are always real -- np.linalg.eigh() is
both the numerically correct choice for a symmetric matrix and
sidesteps this NumPy 2.0 change. The complex dtype leaking downstream
into v/r is what broke the following cells' plot_arrow() (a bare
TypeError: ufunc 'hypot' not supported for the input types) and
np.arctan2() calls.
As printed:
e, v = np.linalg.eig(E)Current toolbox syntax:
e, v = np.linalg.eigh(E)Notebook: chap11.ipynb
Reason: MVTB 2.0.0 split what used to be a single callable stats()
into two: .stats, a dict-valued property (doesn't print), and
.printstats(), a method that prints the same summary the book's
stats() call used to. Same change at 4 call sites.
As printed:
street.stats()Current toolbox syntax:
street.printstats()Notebooks: chap11.ipynb, chap14.ipynb
Reason: MVTB 2.0.0 changed Image.Zeros()/Image.Constant()/etc. to
take image dimensions as a keyword-only size= tuple rather than
positional width/height arguments.
As printed:
canvas = Image.Zeros(1000, 1000, dtype="uint8")
K = Image.Constant(21, 21, value=1/21**2)Current toolbox syntax:
canvas = Image.Zeros(size=(1000, 1000), dtype="uint8")
K = Image.Constant(value=1/21**2, size=(21, 21))Notebook: chap7.ipynb
Reason: RTB's June 2026 xacro/URDF rework removed URDF_read() as a
static method on Robot/ERobot entirely (no deprecation shim -- this
was an architectural change, not a rename) and replaced it with a
module-level function in roboticstoolbox.models.URDF.URDFRobot. The
loading mechanism changed too: RTB no longer bundles raw xacro package
trees like ur_description/ directly, so a full relative xacro path no
longer resolves. Bare robot names ("ur5") now resolve via the
robot_descriptions package instead, matching how the packaged
models.URDF.UR5() class itself loads its data.
As printed:
urdf, *_ = ERobot.URDF_read("ur_description/urdf/ur5_joint_limited_robot.urdf.xacro")
urdfCurrent toolbox syntax:
from roboticstoolbox.models.URDF.URDFRobot import URDF_read
urdf, *_ = URDF_read("ur5")
urdfNotebook: chap14.ipynb
Reason: The bridge dataset has 251 stereo frame pairs; processing all
of them (ORB matching + bundle adjustment per frame) takes several
minutes. Added an optional frame-count argument to
RVC3/examples/visodom.py (processes every frame if omitted) and the
notebook now passes 50 for a much faster run. Increase or drop the
argument to process more frames.
As printed:
%run -m visodomCurrent toolbox syntax:
%run -m visodom 50Notebook: chap15.ipynb
Reason: bdsim's per-clock logged state array was renamed from a
lowercase .x attribute to .X, to match the uppercase convention
used for state-array fields elsewhere in a run's output struct (eg.
the top-level out.x/out.xnames pairing uses lowercase for the
continuous state but clocked/discrete state blocks use the
uppercase form).
As printed:
plt.plot(out.clock0.t, out.clock0.x)Current toolbox syntax:
plt.plot(out.clock0.t, out.clock0.X)Notebook: chap2.ipynb
Reason: RTB used to ship tripleangledemo.py as a standalone
script; it's now roboticstoolbox.demo.tripleangledemo, exposed as
an installed console command ([project.scripts] entry point in
RTB's pyproject.toml) rather than a bare top-level module, so
%run -m tripleangledemo can no longer find it
('tripleangledemo' is not a valid modulename on sys.path). Invoke
it as the installed command directly instead.
As printed:
%run -m tripleangledemoCurrent toolbox syntax:
!tripleangledemoNotebook: chap2.ipynb
Reason: Same change as tripleangledemo above --
twistdemo.py is now roboticstoolbox.demo.twistdemo, an installed
console command rather than a bare top-level module.
As printed:
%run -m twistdemoCurrent toolbox syntax:
!twistdemoNotebook: app.ipynb
Reason: Same change as tripleangledemo/twistdemo above --
eigdemo.py is now roboticstoolbox.demo.eigdemo, an installed
console command rather than a bare top-level module.
As printed:
%run -m eigdemo 1 2 3 4Current toolbox syntax:
!eigdemo 1 2 3 4Checked 2026-10-13 against the released packages in a clean venv (RTB 1.4.4,
MVTB 2.4.0, bdsim 1.4.0, pgraph-python 1.0.0, NumPy 2.5; torchvision 0.29.1
checked by reading its source, not executed). Each of these used to have an
entry here. The printed form still works, at most with a DeprecationWarning,
so a reader doesn't have to change anything. If a future release removes
the alias, the entry should come back.
| Printed | Replacement | Why it still works |
|---|---|---|
ET.eta (§7.1.1.1) |
.param |
RTB keeps eta as a deprecated alias (RTB#553) |
im.image, .A (§10.4.2, §11.1, §11.5.1.2, §12.x, §13.4, §14.x) |
.array |
MVTB 2.0.0 deprecated aliases |
Histogram.plot("ncdf") (§11.2) |
"cdf" |
accepted as a deprecated alias (MVTB#85) |
PointCloud.Read('data/bunny.ply') (§14.7) |
Read('bunny.ply') |
redundant data/ prefix is now stripped (MVTB#92) |
fcn_resnet50(pretrained=True) etc. (§12.1.1.3, §12.1.4) |
weights= |
torchvision still maps pretrained with a warning |
g[1].adjacent() (§I) |
.neighbours() |
pgraph restored adjacent() as the canonical name (2026-10-12) |
Image.rank() (§11.5.2.2) |
.rankfilter() |
deprecated alias |
ImageCollection, ZipArchive (§11.1.2, §12.4.2, §13.2.2, §14.8.x) |
FileCollection, FileArchive |
deprecated subclasses, same arguments |
VideoCamera.grab(), WebCam.grab() (§11.1.3, §11.1.5) |
next(camera) |
deprecated methods call next() |