Kuis AI & Robotika

Jawaban dan penjelasan muncul saat Anda mengikuti kuis.

1. Before testing a robot policy, which boundaries should the team define?
   • Only the desired completion time
   • Only the model parameter count
   • The task, workspace, action limits, and stop conditions
   • Only the wording of a command
2. A robot plan targets the correct location, but its camera misses a person entering the workspace. What should the control layer do?
   • Continue because the plan is syntactically valid
   • Replace physical checks with a text explanation
   • Assume the missing person is outside the workspace
   • Stop or limit motion through the appropriate safety controls
3. Why should a robot policy be evaluated across different lighting, surfaces, and starting states?
   • To guarantee every future condition was tested
   • To avoid collecting outcome measurements
   • To gather evidence about performance under relevant physical variation
   • To remove all need for calibration
4. Which evaluation record adds information beyond task-completion percentage?
   • Only the number of words in the robot’s response
   • Collisions, near misses, recovery time, and operator workload
   • Only the robot’s marketing name
   • Only a count of successful demo videos
5. A policy is moved to different robot hardware. What evidence is needed before assuming its skills transfer?
   • Only evidence that the model file loads
   • Only a matching task name
   • Evaluation for the intended robot embodiment and setting
   • Only a language description of the new robot
6. Which information makes robot trials easier to interpret and reproduce?
   • Only the date of the public announcement
   • Only the number of successful trials
   • Only the operator’s preferred interface
   • Robot, firmware, policy version, calibration, and environment
7. A robot completes a simulated task successfully. What does that result establish about safe operation near unfamiliar people or objects?
   • It proves the robot is safe if the task goal is unchanged
   • It does not establish safety in an unfamiliar physical setting
   • It proves the camera cannot miss an obstacle
   • It removes the need for hardware stop conditions
8. How should a robot’s action path handle a generated command?
   • Restrict command access, validate inputs, and verify the physical outcome
   • Treat any generated text as authorization
   • Let the model disable interlocks when uncertain
   • Assume completion text proves that the motion occurred