Urban sensing is emerging as a significant Wireless Sensor Networks (WSNs) application. In such a scenario, static sensors are sparsely deployed in a urban area to collect environmental information. Sensed data are opportunistically collected by Mobile Sinks (MSs), which can be other sensor nodes attached to cars or buses, or carried by people while they move around the city. Since the contacts between the MSs and the static sensors are infrequent and short, reliable and energy efficient data collection is a primary concern of such applications. To this end, we exploit a hybrid data delivery scheme based on both Erasure Coding (EC) and feedback by the MSs. We provide an optimized implementation, and show by extensive experiments in a real testbed that the proposed approach is feasible, despite the very limited storage and processing resources of commercially available sensor platforms. We also demonstrate that our approach can achieve a high probability of correct data delivery, as well as a high energy efficiency, especially when multiple MSs are in contact with static sensors at the same time.