Demystifying Area AI’s $ 405 M and Genesis AI’s $ 105 M rounds with my lens as a robotics designer
In my last year at Cruise ship (before we got entirely absorbed into General Motors), I came across a construction zone situation that crystallized the essential limitation of present robotics AI: that robots could see isolated aspects but missed out on the huge picture.
Our L 4 assumption system was actually superb at independently identifying cones and barriers (with 90 %+ precision), yet it conveniently missed what those items suggested together A mix of cones and obstacles forming a loosened line commonly signals a shut lane to human vehicle drivers– that we intuitively comprehend its intent to redirect website traffic or block entry. But understanding systems do not presume intent. If even one cone is missed out on or the voids show up just vast sufficient, the system might analyze the scene as a drivable lane.
This type of failure mode sadly had tragic ramifications, as the car might drive straight into an active job zone with building workers, open manholes, heavy machinery or debris, while all at once blocking website traffic behind it up until human safety and security operators interfered.