Quiz#

What is the advantage of using simulation in robot learning?
- A. It’s more cost-effective
- B. It allows for faster than real-time training
- C. It provides access to privileged information
- D. All of the above

The “reality gap” in sim-to-real transfer refers to:
- A. The difference between simulation and real-world conditions
- B. The time it takes to deploy a policy
- C. The cost difference between simulation and real robots
- D. The physical distance between the simulation computer and the robot

Which of the following is not a component of the reality gap?
- A. Approximation errors
- B. Model errors
- C. Unmodelled dynamics
- D. User interface errors

Domain randomization always improves the performance of a policy when deployed on hardware.
- A. True
- B. False

What is the primary purpose of regularization in sim-to-real transfer?
- A. To increase training speed
- B. To create more realistic simulations
- C. To encourage smooth, hardware-friendly behavior
- D. To improve visual rendering in simulations

In the context of privileged information, what does the “teacher” in a teacher-student approach have access to?
- A. Only real-world data
- B. Only simulated data
- C. Both simulated and real-world data
- D. Information not available to the real robot

Which of the following is a benefit of using actuator networks in sim-to-real transfer?
- A. They eliminate the need for real robots
- B. They accurately model complex actuator behaviors
- C. They speed up physical robot movement
- D. They reduce the cost of robot hardware

The goal of visual domain randomization is to achieve photorealism.
- A. True
- B. False

What is a key consideration when balancing domain randomization and real-to-sim approaches?
- A. Always prioritize domain randomization
- B. Focus solely on real-to-sim techniques
- C. Find a balance that creates robust yet effective policies
- D. Ignore both approaches and rely on intuition