Explain the trade-offs between centralized and decentralized decision-making in autonomous vehicle systems.

Example answer

Centralized processing can simplify access to a common world model and coordinated scheduling, but it can concentrate compute, communication or power dependencies. Distributed processing can place work near sensors and isolate some faults, yet adds synchronization, communication, versioning and agreement problems. Neither layout is inherently faster or fault tolerant.

Map the end-to-end deadlines and failure consequences before allocating functions. Specify where authoritative state lives, how stale or conflicting estimates are detected, and which component arbitrates actuator commands. Two independent planning processes should not send incompatible commands without a defined arbitration contract.

Analyze shared resources: processors may be separate while still depending on one clock, power supply, network switch or software defect. Partition memory and execution budgets where required, and include communication and serialization costs in timing tests. The fault-detection and arbitration components also need their own failure analysis.

Evaluate representative workloads, overload and injected faults across subsystem boundaries. Compare observed deadline misses, fault containment, transition behavior, energy and maintenance complexity. Use the architecture whose demonstrated properties fit the function and operating domain; a block diagram alone is not evidence of resilience.