Free NCA-AMAD Physical AI Practice Questions

Correct answer: C. A Jetson AGX Thor onboard computer

C is correct because the third computer is the robot's own runtime computer, Jetson AGX Thor, which performs inference on the robot's sensor data. A is incorrect because a cloud instance is not mounted on the robot and cannot provide real-time onboard inference. B is incorrect because Omniverse and Cosmos on RTX PRO Servers form the second computer, used for simulation and synthetic data. D is incorrect because DGX is the first computer, used to train the robot's AI models.

Correct answer: B. Physical AI acts in the physical world; agentic AI works in digital environments

B is correct because NVIDIA describes physical AI as end-to-end models that perceive, reason, interact with and navigate the physical world, whereas agentic AI operates in digital environments. A is incorrect because both kinds of AI run on accelerated GPU computing; the difference is where the system acts, not the processor type. C is incorrect because image and video generation are still digital outputs; physical AI is defined by perceiving and acting in the real world. D is incorrect because physical AI needs large, diverse training data too, which is exactly why it relies on simulation and synthetic data.

Correct answer: C. A surgical robot that learns to thread needles and place stitches

C is correct because physical AI lets machines perceive, reason and act in the physical world, and NVIDIA cites surgical robots learning to thread needles and stitch as an example. A is incorrect because retrieval over documents is a digital AI workload, not an autonomous machine acting in a physical space. B is incorrect because summarizing documents is a digital task; nothing in it senses or acts on the physical world. D is incorrect because a chatbot works only with text in a digital environment and takes no action in the physical world.

Correct answer: A. LLMs learn from internet data and lack a grasp of 3D space and physics

A is correct because NVIDIA notes that models trained on internet text and images are limited in their grasp of the physical world and its rules; physical AI adds an understanding of 3D spatial relationships and physical behavior and turns sensor input into actions. B is incorrect because many generative models accept images and can be fine-tuned; the issue is that their training data does not teach physical rules. C is incorrect because LLMs are often far larger than robot policies; the gap is in physical understanding, not parameter count. D is incorrect because LLMs run on GPUs as a matter of course; GPU support is not what separates them from physical AI.

Correct answer: D. RTX PRO Servers running Omniverse and Cosmos

D is correct because the second computer runs Omniverse and Cosmos on RTX PRO Servers so developers can test robot policies in digital twins and generate synthetic data before real deployment. A is incorrect because the runtime computer runs the deployed policy on the robot; the team wants to test before any robot runs it. B is incorrect because replaying old logs cannot show how a new policy behaves in new situations, and it is not part of the three-computer pattern. C is incorrect because DGX is the training computer, where frameworks such as PyTorch and TAO pre-train and post-train models; it is not the platform for simulating test scenarios.

Correct answer: B. On an onboard computer such as Jetson AGX Thor inside the robot

B is correct because the runtime computer processes sensor data, reasons, plans and acts within milliseconds on the robot itself, so the safety response does not depend on a network link that can drop. A is incorrect because data center servers suit larger inference workloads without millisecond limits; unreliable Wi-Fi makes them unsafe for collision avoidance. C is incorrect because fixed scripts cannot perceive a worker stepping into the path, so the robot would have no real-time response at all. D is incorrect because every decision would add a network round trip, and any loss of signal would leave the robot unable to react in time.

Correct answer: B. Enough inference per watt to serve several models at once

B is correct because the battery limits power while three models must answer together, and NVIDIA describes the on-robot computer as compact and energy efficient so it can run an ensemble of control, vision and language models in real time. A is incorrect because rendering synthetic data is the job of the simulation computer, not the battery-powered computer on the robot. C is incorrect because large-scale data curation happens off the robot; on board, the priority is real-time inference within a power budget. D is incorrect because pre-training is the job of the DGX training computer in the data center, not the robot's onboard computer.

Correct answer: A. The Metropolis AI Blueprint for video search and summarization (VSS)

A is correct because NVIDIA points to the Metropolis AI Blueprint for VSS to build video analytics AI agents that analyze factory performance and improve worker safety at the edge. B is incorrect because Isaac Lab trains robot policies in simulation; it does not analyze or summarize camera feeds. C is incorrect because GR00T N1 outputs robot actions for manipulation; it is not a video search and summarization workflow. D is incorrect because Mega is for simulating and testing robot fleets in a factory digital twin, not for analyzing live camera video.

Correct answer: D. NVIDIA DRIVE AGX

D is correct because NVIDIA states that the optimized stack and policy can be deployed on Jetson or DRIVE AGX to run embedded in robots and vehicles, and DRIVE AGX is the vehicle option. A is incorrect because RTX PRO Servers run simulation and larger data center workloads, not the onboard driving stack. B is incorrect because DGX is the training computer in the data center; it is not embedded in the vehicle. C is incorrect because CARLA is a simulator for validating an AV stack before deployment, not a computer that runs in the car.

Correct answer: A. Let it sense each part's pose and adjust grip position and force in real time

A is correct because physical AI moves robots from rigid automation to autonomy by sensing, reasoning and acting in real time, such as adjusting grasp strength and position to an object's pose on a conveyor. B is incorrect because duplicating the fixed motion adds capacity but no perception, so rotated parts still fail. C is incorrect because logging helps people see the problem, but the arm itself still cannot perceive or adapt. D is incorrect because a faster fixed motion is still rigid automation and will keep dropping rotated parts.