Blogs

On 21 October 2026, Autodesk released the October update for Fusion, its CAD (computer-aided design) software. One new tool is called AutoTimeline. It is a Tech Preview, which means an early test version. Autodesk says AutoTimeline uses AI to study an imported 3D model and rebuild it as editable features, like extrudes and holes. This matters to every design engineer, because fixing supplier files that cannot be edited often takes hours of extra work.

What happened?

• Date: 21 October 2026, in the Fusion October 2026 product update.
• Company: Autodesk, the maker of Fusion, Inventor and AutoCAD.
• New tool: AutoTimeline, a Tech Preview for users of the Fusion Design Extension.
• What it does: Autodesk says it finds design features in imported geometry and rebuilds them as a parametric timeline.
• Background: Autodesk first showed AutoTimeline at its Autodesk University event in September 2026.
• Also in the update: Autodesk says large assemblies now open, save and edit up to 35% faster.

Why are imported parts hard to edit?

When you model a part in SOLIDWORKS, CATIA, NX or Creo, the software records every step. This list of steps is called the feature tree. Fusion calls it the timeline. For example: draw a rectangle, extrude it 8 mm, cut four holes, then round the edges.

Each step keeps its own dimensions. If you change one number, the whole part updates. This is called parametric modelling.

Now imagine a supplier sends you a STEP file. A STEP file is a neutral 3D format that almost all CAD software can open. But it carries only the final shape: a set of faces and edges. The steps and dimensions are gone. Engineers call this a "dumb solid".

To make one hole bigger, you must push and pull faces by hand. For a big change, you often model the whole part again, which can take days.

How does feature recognition work?

Autodesk has not explained exactly how its AI works inside. But the basic idea of feature recognition (software that finds design features in a shape) is well known. It works roughly like this:

1. Read all faces. The software lists every face and its type: flat, cylinder, cone or curved.
2. Find simple features. A cylinder face that goes into the material is probably a hole. A small curved face between two flat faces is probably a fillet (a rounded edge).
3. Remove them one by one. The software fills the holes and removes the fillets. This is called defeaturing. The shape becomes simpler.
4. Find the base sketch. What remains is often a flat profile pushed in one direction. The software turns it into a sketch plus an extrude.
5. Rebuild in order. The base extrude comes first, then the holes, and fillets near the end. Now each feature has its own editable dimensions.

Simple rules work well for plain shapes. Real parts are harder. For example, was a pocket made with one cut or two? Which order makes the most sense? This is where AI can help. Engineering.com noted that other companies, such as Backflip and CADfit, have built similar tools.

What this means for you

AI tools will save time, but you must still judge whether the rebuilt tree is good. Focus on these skills:

• Design intent: learn why features are built in a certain order. A clean tree is easy to change; a messy one breaks.
• Direct editing: practise tools like move face and offset face. You will still need them for odd shapes.
• Checking: AutoTimeline is still a test version. Always check key dimensions against the drawing after any automatic rebuild.

Try this today: In your CAD software, model an angle bracket: 80 mm wide, 50 mm deep, 50 mm tall and 8 mm thick, with two 8 mm holes in each face. Save it as a STEP file. Open the STEP file as a new part. You will see only one item in the feature tree. Now try to change two holes to 10 mm using direct editing tools. Then rebuild the part by hand, following the five steps above, and time yourself.

Want to learn mechanical design step by step from an industry expert? Join our free live workshop.

Sources: Autodesk Fusion Blog, 21 Oct 2026 · Engineering.com, 15 Sep 2026

On 21 October 2026, Tesla said it has started early production of Optimus, its humanoid robot (a robot with a body shaped like a human). The robots are made on a new line in Fremont, California, in a factory that used to build cars. Tesla hopes to make 20,000 robots every week one day, but says the start will be slow. For a design engineer, this is a good lesson in robot joints, actuators and designing a product that can be made in large numbers.

What happened?

• Date: Tesla shared the news with its July–September 2026 results on 21 October 2026.
• Place: Tesla says early production began on a "high-volume" line in Fremont, California.
• Space: Reports say the old Model S and Model X car lines were removed this summer to make room. The work took 46 days.
• Plan: The Fremont line is reported to be planned for 1 million robots per year.
• Target: Elon Musk says 20,000 robots per week. He expects this to take at least one year.
• Customers: Musk expects first deliveries to outside customers around the end of 2027.
• Hands: Reports say each hand of the production design has 22 degrees of freedom.

Musk also said the design still needs to improve, for example in reliability (working for a long time without breaking). So this is a start, not a finished product.

What is a "degree of freedom"?

A degree of freedom (DOF) is one independent way a part can move. Your elbow bends in one direction, so it has about one DOF. Your shoulder can swing forward, swing sideways and twist, so it has about three.

A robot needs one actuator for each DOF. An actuator is the small "muscle" that makes a joint move. So a hand with 22 DOF needs 22 controlled movements, all packed inside a small space.

How does one robot joint work?

Most robot joints follow the same simple loop.

• Step 1: The controller (a small computer) decides the angle the joint should reach.
• Step 2: An electric motor spins very fast, but its torque (turning force) is low.
• Step 3: A gearbox slows the spin and multiplies the torque. A 10:1 gear ratio gives about 10 times more torque.
• Step 4: The joint turns and moves the arm or finger.
• Step 5: A sensor measures the real angle and sends it back. The controller corrects the motor until the angle is right.

Back in January 2025, Musk said Tesla could not find any ready-made motors, actuators or sensors that worked for a humanoid hand. Tesla says it designed its own parts.

Why is making many robots harder than making one?

Building one robot in a lab is a research job. Building thousands every week is a design job. Every part must be easy to make, easy to assemble and the same every time. Engineers call this design for manufacturing (DFM). A humanoid has many moving joints, so small errors in each part can add up. Heat from motors, wear in gears and the weight of the battery also matter.

What this means for you

If you want to work in robotics or automation, learn these skills:

• Mechanisms and gears: how gear ratios change speed and torque.
• Motor sizing: choosing a motor that gives enough torque without being too heavy.
• Tolerance stack-up: checking how small size errors add up across many parts.
• Assembly modelling: building joints with proper constraints in SOLIDWORKS, CATIA, NX or Creo.

Try this today: A robot arm holds a 1 kg mass at 0.5 m from the elbow. The weight is 1 × 9.81 = 9.81 N. The torque at the elbow is 9.81 × 0.5 = about 4.9 N·m. If your motor gives 0.5 N·m, you need a gear ratio of at least 10:1 (ignoring the arm's own weight and friction). Now model a simple two-link arm in your CAD software with two revolute (hinge) joints. Count its degrees of freedom and check that it moves the way you expect.

Want to learn mechanical design step by step from an industry expert? Join our free live workshop.

Sources: Not a Tesla App, 22 Oct 2026 · ETV Bharat, 22 Oct 2026 · TechCrunch, 29 Jan 2025

Making a new car usually takes 4 to 5 years, from the first idea to the factory. On 21 October 2026, General Motors (GM), one of the biggest car companies in the world, said it wants to do this in about 2 years. The big idea is simple: test the car inside a computer first, and build real parts later. In this post we explain how it works in easy words, and what it means for you as a mechanical design engineer.

What did GM say?

• At an event called GM Forward in New York, GM said a new car normally takes 4–5 years from idea to production. Its new goal is about 2 years. GM did not say when every new model will reach this speed.
• For its next electric car battery, GM tested more than 300,000 battery designs on the computer before it made the first real prototype.
• GM says its coming electric cars will be about 15% lighter and 25% more aerodynamic (air moves around them more easily).
• Why the hurry? A study by the consulting firm AlixPartners says Chinese car makers bring new cars to market about twice as fast as older car companies.

The old way: one step after another

In the old way, engineers design a part, then make a real sample of it (a prototype), then test it. If the test fails, they change the design, make a new prototype and test again. Every loop can take weeks or months. Also, one team often has to wait for another team to finish first.

The new way: test in the computer first

Today, engineers can build a full digital model of the car and test it with software. This is called simulation. The computer can check if a part is strong enough, how the car will behave in a crash, and how air flows around the body. If something fails, the engineer changes the CAD model and runs the test again. No metal is cut.

AI is making these tests much faster. In a June 2026 report by IEEE Spectrum, GM said a front crash calculation that used to take about 15 hours now takes under one minute with an AI method. GM also used generative design for small brackets in the Corvette: the computer suggested a shape that looks like tree roots, and the new part came out lighter and stiffer than the old one. GM says it has done a fast program before: the GMC Hummer EV went from first design to showroom in about 2 years.

Because the tests are fast, different teams (body, battery, electronics, software) can work on the same digital car at the same time. This is called concurrent engineering, and it is a big reason the total time can drop by half.

Does this mean no more real testing?

No. A simulation is only as good as the data you give it. Wrong material values or wrong loads give wrong answers. Safety approval still needs real crash tests, and real cars are still driven for many kilometres. The change is that companies now build fewer prototypes and find most problems early, when they are cheap to fix.

What this means for you

• CAD alone is not enough. Companies want design engineers who also understand how their part will be tested: loads, supports, materials and stress.
• Basics still matter. A part that passes in the computer must still be made in a real factory. Design for manufacturing and GD&T are still key skills.
• Learn to read results. Know what stress, deflection (bending) and factor of safety mean, and notice when a result looks wrong.

Try this today: Model a simple L-shaped bracket in your CAD software. Most CAD tools (for example SOLIDWORKS, CATIA, NX and Creo) have a simulation add-on. Fix one face, put a load on the other end, and run the study. See where the stress is highest. Now remove some material where the stress is low and run it again. You just did a small version of what GM does on a full car.

Want to learn mechanical design step by step from an industry expert? Join our free live workshop.

Sources: CNBC, 21 Oct 2026 · IEEE Spectrum, 17 Jun 2026