Welcome, sci-fi realists…

This week in “wait, robots can replace that, too?”: temple elephants. This time, it’s the elephants used in Hindu temple displays in India. 

A mom-and-pop backyard workshop owned by a mechanical engineer named Prasanth Prakashan is building robotic elephants for a good cause. 

Robot elephants that squirt water from their trunks might actually be the heroes no one knew they needed. By replacing live elephants in even some temple ceremonies, these bots could help keep real elephants out of captivity and with their herds. 

Humanoid robots mimic human movements (duh, that’s why we’re here, right?) But, having giant land animals mimicked by mechanical parts to protect real ones is a practical idea we hadn’t considered. 

A big difference here is that for animatronic elephants, fiberglass, iron, and rubber are enough to get the job done. They don’t need the physical AI systems humanoids require, because the job here isn’t stocking shelves or folding laundry. It’s looking majestic and occasionally spraying water. It makes us wonder if someone plans to do this with whales next. 

Anyway, here's what you need to know about robotics right now:

London-based robotics company Humanoid does exactly what its title suggests. But, it’s how the company plans to produce its humanoids at scale that should grab everyone’s attention. Humanoid just introduced an AI framework called KinetIQ Ascend, which will train its humanoids using reinforcement learning. 

Let’s back up for a second. For those who do or don’t know, reinforcement learning was already a method for training AI software for years, where an AI model is “rewarded” for the right actions and steered away from the wrong ones. Applying it to ensure industrial robots are actually helping on a factory floor or warehouse is genius. 

How does it work in this new context? KinetIQ refines the behavior of humanoids as they perform specific tasks over time with this type of reward model so they learn to operate as intended. One such task is picking steel bearing rings from a bin and placing them on a conveyor belt as mentioned in the release. 

A nifty video from Humanoid’s YouTube channel (scroll down a smidge) differentiates the company’s reinforcement learning approach from “imitation learning,” where months of training data on human demonstrations would be collected for humanoids to imitate. 

In the video, Humanoid’s, well, humanoids are seen doing three specific tasks after being trained on KinetIQ: tote handling, object picking, and machine feeding (in this case, placing those steel bearing rings onto conveyor belts). 

What this likely means for the future: Humanoid wants the world of robotics to know there are more efficient training methods than data collection. They point the way to days-long training as opposed to months of building datasets. Expect more companies to come up with their own proprietary methods not named “KinetIQ Ascend.”

Our FYI read: Because there is now a humanoid race going on and speed matters in a race, KinetIQ Ascend’s success will spur more robotics companies to try reinforcement learning to save time.

If you didn’t know, a new use case for autonomous mobile robots (AMRs), aside from anything they can already do in an Amazon warehouse, is apparently…moving railcars! Wait, huh? 

The clip below from X/Twitter shows one of KUKA’s signature orange robots moving a railcar in a German Siemens facility. The specific AMR product line in question is OmniMove, which can be seen carrying who-knows-how-many exact tons. Clearly, it’s a step up from the typical AMR that transports food pallets from one side of a warehouse to another. Expect more heavy lifting clips like these in the future.

Check out this OmniMove in action:

This week’s installment is about a humanoid out of a workplace in Indonesia appearing to approach office staff and perform a programmed roundhouse kick right near them. Before you freak out, this was actually a staged event meant to test the limits of mobility and control for humanoids and ensure they don’t malfunction. The bot’s handlers initially posted the footage on TikTok purposely to turn heads. 

What this little stunt shows is that it only works under perfectly controlled conditions and might fail if there’s even a slight deviation or software malfunction. So, we don’t recommend signing up as a willing participant for such a stunt. A humanoid lunging at you with its martial arts knowledge will be terrifying regardless of whether or not it actually intends to hurt you.

Teaching a robot kung fu sounds cool and all, but it doesn’t take much foresight to see all - the - ways it could be problematic.

As it turns out, the actual bad behavior was by observers on social media spreading misinformation and claiming the bot was a “rogue agent” attacking office workers. Context changes everything. 

If you want to catch up on today’s top AI tools, learn how to actually use AI at your job, or build skills that help you land a better one, this is for you. What you see today is just the beginning. We’ll be adding new courses as fast as we can make them, including personalized course paths, breaking tool-of-the-week lessons, and more based on your feedback.

Get started with your 7-day free trial. After that, early bird pricing is $499/year. Standard pricing is $999/year. Same library, same access; the only difference is when you click. → Browse the course library.

P.S. If your company reimburses professional development, reply to this email and we’ll send a one-page brief you can forward to your manager.

P.S. If you haven’t already… follow us on Youtube!