Topics
Latest
AI
Amazon
Image Credits:Covariant
Apps
Biotech & Health
Climate
Image Credits:Covariant
Cloud Computing
Commerce Department
Crypto
Image Credits:Covariant
Enterprise
EVs
Fintech
Fundraising
Gadgets
back
Government & Policy
ironware
Layoffs
Media & Entertainment
Meta
Microsoft
concealment
Robotics
Security
societal
Space
inauguration
TikTok
shipping
speculation
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
newssheet
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
The UC Berkeley spinout says its new AI platform can help robots think more like people
Covariantthis workweek announced the launch of RFM-1 ( Robotics Foundation Model 1 ) . Peter Chen , the co - founder and CEO of the UC Berkeley artificial tidings spinout tells TechCrunch the platform , “ is basically a enceinte language manakin ( LLM ) , but for golem language . ”
RFM-1 is the result of , among other things , a monolithic trove of data point collected from the deployment ofCovariant ’s Brain AI political platform . With client consent , the startup has been building the robot eq of an LLM database .
“ The visual sense of RFM-1 is to power the billions of automaton to come , ” Chen sound out . “ We at Covariant have already deployed lots of robots at warehouses with success . But that is not the limit of where we want to get to . We really want to power automaton in manufacture , food processing , recycling , Agriculture Department , the service industry and even into people ’s homes . ”
The platform launch as more robotics firms are discourse the futurity of “ oecumenical intention ” systems . The sudden onslaught of humanoid robotics firms like Agility , Figure , 1X and Apptronik has represent a polar office in that conversation . The form factor is particularly suited to adaptability ( much like the mankind on which it ’s pose ) , though the robustness of on - board AI / software systems is another question entirely .
For now , Covariant ’s software is largely deploy on industrial automatonlike arms doing a multifariousness of intimate warehouse tasks , including chore like bin picking . It is n’t presently deployed on android , though the ship’s company is promising some grade of hardware agnosticism .
“ We do like a lot of the work that is happening in the more world-wide purpose robot computer hardware space , ” say Chen . “ Coupling the intelligence flexion item with the hardware flection point is where we will see even more explosion of automaton applications . But a lot of those are not fully there yet , especially on the hardware side . It ’s very hard to go beyond thestaged video . How many multitude have interacted with a humanoid in someone ? That tell you the level of maturity . ”
Covariant does n’t , however , shy away from human comparisons when it get along to the role RFM-1 plays in golem ’ decision - making process . Per its press material , the platform , “ provides robots the human - like ability to reason , representing the first prison term Generative AI has successfully given commercial-grade robots a abstruse understanding of language and the physical world . ”
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
This is one of those realms where we have to be careful with claims , both in terms of comparisons to abstract — or even philosophic — concepts and their real real - human race efficaciousness over time . “ Human - same ability to reason ” is a broad - wholesale conception that means a lot of different thing to a lot of different people . Here the notion apply to the system ’s power to work real - humankind data and determine the safe line of activity to execute the chore at helping hand .
thing can break down quickly , however , with even the humble deviations . Say the object is n’t placed exactly properly on the conveyor belt belt , or there ’s been an registration to lighting that impacts on - board cameras . These sorts of differences can have a Brobdingnagian shock on the robot ’s power to execute . Now imagine trying to get that robot to work with a new part , new material or even do a wholly dissimilar task . That ’s even harder .
This is the point where programmers traditionally step in . The golem must be reprogrammed . More often than not , someone from outside the factory flooring get into the picture . This is a bragging drainpipe of resources and time . If you want to nullify this , one of two thing needs to happen : 1 ) People working on the level need to find out computer code or 2 ) You require a new , more natural method acting for interacting with the robot .
While it would be great to do the former , it seems improbable that companies will be unforced to endow the money and hold off the necessary meter . The latter is on the nose what Covariant is attempting to do with RFM-1 . “ ChatGPT for robots ” is n’t a sodding analogy , but it ’s a reasonable shorthand ( specially in light of the founders ’ connection to OpenAI ) .
From the client ’s point of position , the program presents as a text field , much like the current iteration of consumer - facing reproductive AI . Input a text bidding like , “ pick up the orchard apple tree ” by typing or representative , and the scheme uses its training data point ( cast , color , size , etc . ) to identify the object in front of it that most closely matches that description .
RFM-1 then generates video outcomes — in essence simulations — to fix the expert course of natural action using retiring training . This last scrap is similar to how our brain work out the likely outcomes of an action prior to executing .
During a live demo , the system reacts to inputs like “ cull up the red physical object ” and even the more semantically complex , “ pick up what you put on your feet before you put on your shoes , ” which make the robot to aright piece up the apple and a pair of sock , respectively .
A passel of grownup idea are tossed around when discussing the system ’s hope . At the very least , Covariant has an impressive parentage among its founders . Chen meditate AI at Berkeley under Pieter Abbeel , his Covariant co - laminitis and chief scientist . Abbeel also became an other OpenAI employee in 2016 , a calendar month after Chen fall in the ChatGPT firm . Covariant was founded the following yr .
Chen say the company expects the Modern RFM-1 platform will work with a “ majority ” of the hardware on which Covariant software is already deployed .
