RAPID is an intentional DTN routing protocol that optimizes specific routing metrics such as worst-case delivery delay or the fraction of packets delivered within a deadline. It treats DTN routing as a resource allocation problem, translating routing metrics into per-packet utilities that determine how packets should be replicated. RAPID replicates packets opportunistically until a copy reaches the destination. It uses a control plane to track network resources and exchange metadata among nodes. RAPID's control channel allows nodes to share information about packet replicas and transfer opportunities, improving routing performance. RAPID was deployed on a vehicular DTN testbed consisting of 40 buses, covering a 150 square-mile area around Amherst, MA. The deployment included 58 days of performance traces. RAPID significantly outperformed existing routing protocols for several metrics, including average delay, worst-case delay, and packet delivery within a deadline. For small loads, RAPID performed within 10% of optimal performance. RAPID's control channel, which exchanges metadata, was found to be efficient, with metadata overhead accounting for less than 0.02% of total available bandwidth. RAPID was evaluated using a trace-driven simulator, which showed performance results within 1% of real measurements with 95% confidence. RAPID was compared to four existing routing protocols and random routing. The results showed that RAPID significantly outperformed these protocols for all metrics. For example, in trace-driven experiments under moderate-to-high loads, RAPID outperformed the second-best protocol by about 20% for all three metrics, while also delivering 15% more packets for the first two metrics. With a priori mobility information and moderate-to-high loads, RAPID outperformed random replication by about 50% for high packet loads. RAPID was also compared to an optimal protocol, showing that it performed within 10% of optimal for low loads. The protocol's performance was evaluated using three metrics: minimizing average delay, minimizing worst-case delay, and maximizing the number of packets delivered within a deadline. The results showed that RAPID significantly outperformed four other routing protocols. RAPID's performance was also evaluated using synthetic mobility models, showing consistent improvements across different scenarios. RAPID's design includes a selection algorithm, an inference algorithm, and a control channel. The selection algorithm determines which packets to replicate at a transfer opportunity based on their marginal utility. The inference algorithm estimates the utility of a packet based on the routing metric. The control channel propagates the necessary metadata required by the inference algorithm. RAPID's performance was validated using a trace-driven simulator and real-world deployment on a vehicular DTN testbed. The results showed that RAPID significantly outperformed existing routing protocols for various metrics, including average delay, worst-case delay, and packet delivery within a deadline.RAPID is an intentional DTN routing protocol that optimizes specific routing metrics such as worst-case delivery delay or the fraction of packets delivered within a deadline. It treats DTN routing as a resource allocation problem, translating routing metrics into per-packet utilities that determine how packets should be replicated. RAPID replicates packets opportunistically until a copy reaches the destination. It uses a control plane to track network resources and exchange metadata among nodes. RAPID's control channel allows nodes to share information about packet replicas and transfer opportunities, improving routing performance. RAPID was deployed on a vehicular DTN testbed consisting of 40 buses, covering a 150 square-mile area around Amherst, MA. The deployment included 58 days of performance traces. RAPID significantly outperformed existing routing protocols for several metrics, including average delay, worst-case delay, and packet delivery within a deadline. For small loads, RAPID performed within 10% of optimal performance. RAPID's control channel, which exchanges metadata, was found to be efficient, with metadata overhead accounting for less than 0.02% of total available bandwidth. RAPID was evaluated using a trace-driven simulator, which showed performance results within 1% of real measurements with 95% confidence. RAPID was compared to four existing routing protocols and random routing. The results showed that RAPID significantly outperformed these protocols for all metrics. For example, in trace-driven experiments under moderate-to-high loads, RAPID outperformed the second-best protocol by about 20% for all three metrics, while also delivering 15% more packets for the first two metrics. With a priori mobility information and moderate-to-high loads, RAPID outperformed random replication by about 50% for high packet loads. RAPID was also compared to an optimal protocol, showing that it performed within 10% of optimal for low loads. The protocol's performance was evaluated using three metrics: minimizing average delay, minimizing worst-case delay, and maximizing the number of packets delivered within a deadline. The results showed that RAPID significantly outperformed four other routing protocols. RAPID's performance was also evaluated using synthetic mobility models, showing consistent improvements across different scenarios. RAPID's design includes a selection algorithm, an inference algorithm, and a control channel. The selection algorithm determines which packets to replicate at a transfer opportunity based on their marginal utility. The inference algorithm estimates the utility of a packet based on the routing metric. The control channel propagates the necessary metadata required by the inference algorithm. RAPID's performance was validated using a trace-driven simulator and real-world deployment on a vehicular DTN testbed. The results showed that RAPID significantly outperformed existing routing protocols for various metrics, including average delay, worst-case delay, and packet delivery within a deadline.