Booster T2：2070 TFLOPS 让机器人本地推理更复杂

Peak TFLOPS don’t matter when I’m building something.

What I care about is how much of that I can spend on inference while the control loop keeps going.

Booster T2’s huge 2070 TFLOPS could really expand that inference budget. Means much more complex vision and policy models to stay in the loop.

My willingness to use policies will change if @boosterobotics T2 allows me to run more complex perception models on-device without exceeding my latency budget.

Local inference also removes network delay, connection loss and the uncertainty of sending critical decisions off the robot.