Intercept Success Rate (ISR) Metric

• Intercept Success Rate (ISR) is a metric that quantifies the feasibility of kinematic interception by evaluating if an interceptor can reach predicted positions within its physical limits.
• ISR is computed as the fraction of prediction trials where the minimum intercept time, derived using a bang–bang control strategy, meets or beat the available time budget.
• Empirical results demonstrate that advanced methods, such as P2P, can achieve approximately 60% interception feasibility, highlighting significant improvements over baseline approaches.

The Intercept Success Rate (ISR) quantifies the feasibility of kinematic interception in autonomous drone pursuit. It directly measures, under strict maximum velocity and acceleration limits, the fraction of trajectory predictions that an interceptor can physically reach within a defined time horizon. Originating in the context of open-world drone chasing, ISR enables rigorous comparison of prediction methods with respect to actionable pursuit, as opposed to purely positional accuracy. The metric is designed to evaluate whether predicted future drone locations are not only accurate but also achievable given the interceptor’s real-world motion constraints (Oruganti, 27 Jan 2026).

1. Formal Definition and Theoretical Foundation

ISR is grounded in a bang–bang minimum-time intercept formulation where the interceptor’s control strategy switches between maximal acceleration and maximal speed:

• For a given Euclidean distance $d = \|\hat{p} - p_0\|$ (from interceptor origin $p_0$ to predicted point $\hat{p}$), the minimum intercept time $t_\mathrm{reach}(d)$ is

$$t_\mathrm{reach}(d) = \begin{cases} \sqrt{\frac{2 d}{a_\mathrm{max}}} & \text{if } d \le d_c \ \frac{v_\mathrm{max}}{a_\mathrm{max}} + \frac{d - d_c}{v_\mathrm{max}} & \text{otherwise} \end{cases}$$

where $v_\mathrm{max}$ and $a_\mathrm{max}$ denote maximum speed and acceleration, with the critical distance $d_c = {v_\mathrm{max}^2}/{2a_\mathrm{max}}$ delineating the regime transition.

ISR is computed over $N$ prediction trials as the mean of binary feasibility evaluations: $$\mathrm{ISR} = \frac{1}{N} \sum_{i=1}^N \mathbb{I}\left[ t_\mathrm{reach}(\|\hat{p}_i - p_0\|) \leq t^*_i \right]$$ where $t^*_i$ is the time-budget (prediction horizon offset) for trial $i$, and $\mathbb{I}$ denotes the indicator function.

ISR thus directly encodes pursuit feasibility as a scalar fraction of reachable predictions under physically plausible motion constraints (Oruganti, 27 Jan 2026).

2. Operational Constraints and Evaluation Protocol

The ISR metric is parameterized by the interceptor’s fixed kinematic and temporal limits:

• Maximum speed: $v_\mathrm{max} = 15$ m/s
• Maximum acceleration: $a_\mathrm{max} = 5$ m/s²
• Critical distance: $d_c = v_\mathrm{max}^2 / (2a_\mathrm{max}) = 22.5$ m
• Prediction horizon: $H = 20$ frames (dataset-specific frame rate)
• For each prediction, the allotted time $t^*_i$ equals the elapsed time between the interceptor’s start and the predicted frame.

This setup models a constrained, physically realistic interceptor and ensures that ISR reflects genuine actionability rather than theoretical optimality unconstrained by motion limits. Each test sequence consists of individual "trials" where the system predicts the future target position and ISR tallies the binary feasibility outcome for each (Oruganti, 27 Jan 2026).

3. Success Criteria and Binary Feasibility

Success in ISR is strictly defined: a prediction is successful if the calculated minimum-time intercept $t_\mathrm{reach}$ is less than or equal to the available time $t^*_i$ for that prediction. If so, the trial contributes 1 to the average; otherwise, it contributes 0. There is no partial credit or further weighting.

As a result, ISR is directly sensitive to both prediction accuracy and physical feasibility. Predictions that are spatially accurate but temporally unreachable due to kinematic limits do not improve ISR. Conversely, trivially reachable predictions (e.g., stationary targets coincident with $p_0$) artificially inflate ISR without meaningful pursuit value—a property exposed in specific baselines (Oruganti, 27 Jan 2026).

4. Computation over Drone Pursuit Sequences

The computation process for ISR is as follows:

• For each of $N$ trials, predict the target’s future position $\hat{p}_i$,
• Compute $d_i = \|\hat{p}_i - p_0\|$ and $t_\mathrm{reach}(d_i)$,
• Compare $t_\mathrm{reach}(d_i)$ to the corresponding $t^*_i$,
• Assign a binary indicator (1 or 0) for each successful or failed trial,
• ISR is the arithmetic mean over all trials.

No special handling (such as clipping or replacement) is applied to failed attempts; they simply result in a $0$ outcome. This Boolean structure focuses measurement on genuinely actionable trajectory predictions (Oruganti, 27 Jan 2026).

5. Empirical Performance and Comparison to Baselines

ISR has been evaluated on the Anti-UAV-RGBT dataset across 8,092 prediction trials, benchmarking diverse baseline and advanced methods:

Method	ADE (px)	FDE (px)	ISR	Drone-Acc
Frame-based	261.07	261.73	1.000	0.000
Tracking-only	122.45	52.53	0.001	0.000
Naïve Velocity	122.83	53.24	0.001	0.000
P2P (Ours)	28.12	41.14	0.597	1.000

Key outcomes:

• The P2P method improves trajectory accuracy (ADE) by 77% relative to tracking-only baselines (from 122.83 to 28.12 pixels).
• Pursuit feasibility, as measured by ISR, shifts from near zero (0.001; "99.9% infeasible") to 0.597, corresponding to approximately 60% of predictions being physically actionable—a 597-fold increase.
• The frame-based baseline achieves ISR = 1.000 by predicting stationary targets at $p_0$, a degenerate case yielding zero drone-classification accuracy and no effective pursuit guidance (Oruganti, 27 Jan 2026).

6. Insights, Limitations, and Potential Extensions

ISR highlights the distinction between prediction accuracy and pursuit actionability: minimizing displacement error (ADE) does not ensure feasible interception unless kinematic constraints are explicitly considered. Temporal motion pattern representation (e.g., through 8-D motion tokens) enables both accurate drone discrimination and physically feasible pursuit without external cues.

Applicability of ISR spans any system requiring real-world actionable predictions, as it quantitatively measures how many predictions can be operationalized under strict actuator constraints.

Identified limitations and suggested extensions include:

• Fixed interceptor parameters; ISR could be tailored for heterogeneous vehicle capabilities or adapted dynamically.
• Current evaluation is dataset-specific (Anti-UAV-RGBT); broader cross-domain validation is warranted.
• Only single-target scenarios are addressed; extension to multi-target interception and real-time control integration remains an open challenge.
• Real-world deployment may require seamless integration with onboard perception and control modules, and may necessitate safety constraints beyond bang–bang minimum-time interception (Oruganti, 27 Jan 2026).

ISR thus offers a principled, physically grounded metric for pursuit feasibility, facilitating discriminative comparison of open-world trajectory prediction methods where kinematic realizability is central.