We’re making Fabricator - a platform that lets you iterate hardware in simulation and then directly deploy the learnt behaviour into the real world.
Currently a lot of questions get answered by slow iterations in the real life.
How many degrees of freedom on which mechanism (robotic arm, gantry, scissor lift…) will give us the required reach and precision at a reasonable price? What sensors should we use and where should we place them to give us the required accuracy in the dusty, dark or went environments that the robots might experience? How much material and batteries should the robot carry, how large should it be? How will all these choices affect the time taken to operate the robots, how many robots one person can manage and general worksite scheduling?
Fabricator will answer all of them within days for you so that you can skip many years of development or go directly to the final prototype
Through text and code you will specify the task and the environment. You would be seeing what designs and behaviours that leads to and tweak it. For each one you would get metrics like how fast it can work, how many a single operator can use, how much coverage they do autonomously, how expensive they are to manufacture, how reliable they are etc. When you’re happy with the selection of robots and how they’re behaving, you would review the best few designs in detail and pick which ones you’d like to go with. Fabricator would directly provide you with complete designs ready for manufacturing, and complete software that can be loaded on the robots.
↳ intended UI for Fabricator
↳ you would see produced designs and behaviours in real time
What can Fabricator do for you now?
By turning your power tools into robots, you can often increase their value more than 10x and reach a much larger market as labour shortages in construction, agriculture, landscaping and others are limiting your reach.
As a pilot, we would explore 100+ designs in Fabricator, so that you can directly go for a design that your customers will love. (mecanum wheels vs tracks, robotic arm vs a gantry, base station vs carrying everything on the robot) High level questions like this are currently hard to answer without spending years of development but can make a 2+x difference in the quality of the robot.
Together we would then look at the 10 best designs, look through multi hour videos to identify problems before spending any engineering effort, and settle on the best one. After that we would support you in bringing that design to reality. With our experience in bringing grouting robots from idea to market in record time, we know what it takes to make a robot succeed.
By automating or partially automating your machinery, you can get your machines to work faster and more efficiently and can increase your market that is being limited by labour shortages in construction, mining, agriculture, forestry and others.
As a pilot, we could 1) find the best possible new designs that using smarter controls could do the work more accurately or allow automation to begin with (precision agriculture, landscaping…), 2) create low level controls that would allow your machinery to work more precisely, faster or in more complex terrains (driving in complex forest terrain, accounting for the load and terrain to drive faster where possible, enabling high level skills like flattening an area, moving a pile or reshaping a bush to allow easier and faster teleoperation).
Together we would then look at the 10 best approaches, look through multi hour videos to identify problems before spending any engineering effort, and settle on the best one. After that we would support you in bringing that design to reality. With our experience in bringing grouting robots from idea to market in record time, we know what it takes to make a robot succeed.
Many product questions can be answered by generating videos
As a pilot, we would explore any modifications you are considering making to your robot to identify they would have on efficiency and coverage for the complete task in realistic environments. For changes like switching to mecanum wheels it is difficult to estimate it’s impact, it is hard to predict what will be the best placement for sensors etc.
Soon we’ll also be able to provide generated control policy that will zero shot work on real robots. Initially this will be great for demos of new robots and testing new designs, eventually it will be good enough for real deployments.
As a pilot, we would find a step in your design process and provide software that would allow you to do that step 10x faster every time.
An example could be software that would design a robot able to move or sort a specific set of items. It would automatically select and configure arms, grippers and accessories used to guarantee the design will be collision free, fast enough and minimizing pneumatic or other control lines for easier assembly and programming. You would only need to provide models of the objects handled and you would get a list of best configurations within hours. You can show your clients the solution on the same day and adapt to any feedback quickly.
From there we would gradually expand it to allow you to make complex robot cells in hours, and eventually full automated lines in hours.
Set up a pilot and speed up your time to market 5x
This isn’t just a simulation solution, it’s accelerating your development 100x
Design automation has now become possible
Reinforcement learning and robotics simulation has been improving a lot recently
In 2019 it became possible to train robots in simulation and deploy them in reality, now it’s possible to train it in a few minutes on a gaming GPU
We’re building on that. Our platform explores different designs and for each one generates optimal controls. From that we can tell how well that design performs the complete task. And even know all the forces, vibrations etc. it will experience over its lifetime so that we can do component level simulations very accurately.
Specifying the task and environment is very simple, since the environment usually doesn’t need to have much fidelity. The whole design process using our platform should generally be doable in under a week.