John Macdonald

Head of AI at Bonsai Robotics

[John Macdonald @jnmacdnld]

proud to have our work be featured on this week in startups! jump to the 8:13 mark to see the amiga max

[@jason @Jason]

Why robots still can’t pick a strawberry https://t.co/dpDwiZl1ZM

[John Macdonald @jnmacdnld]

@mihuq not ready to share details yet! but I would say, with enough high quality data and the right training recipe, these models can be trained *without* frontier lab level resources

[John Macdonald @jnmacdnld]

these orchards are not real, they were generated by the bonsai world model. why are we betting on a world model?

robotaxi companies can create a detailed HD map and digital twin of a city like San Francisco and use it to train their models on all kinds of diverse scenarios. this is expensive, but you can split the cost across many customers all served by that map. in agriculture, where every customer has their own farm, and therefore their own map, this approach breaks completely. it doesn’t scale at all. and,
unlike robotaxis, there is no easy access to a realistic test environment like public roads, so testing becomes a huge bottleneck. I believe this has held back robotics in agriculture for many years, killing many promising agtech startups.

all of this changes with the bonsai world model. by training on years of high-quality real world data, we are starting to be able to generate diverse, realistic and geometrically consistent 3D farm environments without a painstaking 3D reconstruction process that doesn’t scale.

more to come soon! we are just getting started 😉

[Danny Bernstein 🍇🥬 @bernsteinres]

a @bonsairobotics robot applying snail bait in citrus orchards in renmark, australia.

by itself

stacks work between young and mature blocks and handles the commute between them autonomously

[John Macdonald @jnmacdnld]

Lots of people ask when the “ChatGPT moment for robotics” will come. I would argue that it will never come, because robot adoption is happening slowly, all the time. You just don’t know it. It’s very possible you’ve already eaten almonds harvested by Bonsai Intelligence!

[John Macdonald @jnmacdnld]

I love how practical the amiga max is. unlike manually driven tractors, it doesn't have sleek aerodynamic curves. that's because its a piece of autonomous equipment, not a vehicle you're meant to look at every day from inside the cab. it has bulletproof autonomy, and a sleek app

[Danny Bernstein 🍇🥬 @bernsteinres]

this is a computer that drives through a vineyard and sprays with surgical precision. Bonsai Robotics Amiga Max. hard tech is beautiful.

[John Macdonald @jnmacdnld]

learn all the things 🚀

[Deepak Pathak @deepakpathak]

Even after 4yrs of locomotion research, we keep getting surprised by how far we can push the limits of legged robots! We report a major update 🚀🤖

Extreme Parkour: extremely long & high jumps, ramp, handstand, etc. all with a single neural net!

https://t.co/VQSrOOoDHU

🧵(1/n)

[John Macdonald @jnmacdnld]

if true, one of the more interesting impacts of autonomous vehicles for sure

[Stefan Bodnarescu @stefan_bod]

@newscientist @paulg We need to move fast

Autonomous cars are coming and this will decrease the number of organs available for transplant by 33%

[John Macdonald @jnmacdnld]

and same for the idea of general purpose autonomy! once ridiculed as crazy to think about, now widely accepted as not only possible but inevitable

[John Macdonald @jnmacdnld]

its wild to see how quickly the conventional wisdom on doing robot autonomy fully with deep learning has gone from “nah that will never work” to “of course it works”

[John Macdonald @jnmacdnld]

@faraz_r_khan @simonkalouche nah man you’re overthinking this, it’s the 4th july, nobody is going to complain about a flag

[John Macdonald @jnmacdnld]

@Azadux cool capture pattern! do you generally see better results with orbits as opposed to a grid with obliques?

[John Macdonald @jnmacdnld]

@johnjnay @MSFTResearch agreed. surprisingly few people understand the implications of this. that, or everyone is too scared to talk about it

[John Macdonald @jnmacdnld]

when I was in 1st grade I actually used to solder paperclips to blank PCBs because I read in a book if you wanted to make robots you had to be good at soldering and it suggested this method to practice . crazy how different things are now

[John Macdonald @jnmacdnld]

@robfreeborn my personal conclusion has always been that masculinity (and femininity) are a waste of everyone’s time and you should just do what makes you happy and doesn’t hurt other people

[Orangewood @OrangewoodLabs]

we’re building AI-powered robots - that makes us twice as dangerous

[John Macdonald @jnmacdnld]

@stbearman @d_bau13 @Cruise I will add that I ride in the Cruise driverless cars all the time, and encounter them a bunch while walking around the city too. I feel far safer around the Cruise driver than normal drivers. I have nearly been hit by normal drivers so many times, Cruise would never.

[John Macdonald @jnmacdnld]

@stbearman @d_bau13 @Cruise Please consider that if a normal person had been driving during this minor fender bender, nobody would care. Especially by taxi/rideshare driver standards, the Cruise driver is an outright saint. Hyping up every small misstep the Cruise driver makes does not make anyone safer.

[John Macdonald @jnmacdnld]

@random_walker counterpoint: an LLM OS can understand what it’s being asked to do, and refuse to do certain unethical things. Current OSes have no such ability, with results like this:

https://t.co/Xq02pvoif2