There is a moment in every intercept engagement when the math either closes or it doesn't. A drone doing 28 m/s on a straight inbound track at 150 meters altitude presents a window measured in low single-digit seconds. If your intercept geometry hasn't been solved before the engagement trigger fires, the round goes nowhere useful. This is the core kinematic problem in counter-drone defense, and it's harder than it looks from the outside.
What follows is our working understanding of that problem — how we frame it, which variables dominate, and why certain architectural choices in a kinetic intercept system are not really choices at all once you accept the physics.
Closing Velocity and Why It Dominates Everything Else
The most immediate number in any intercept geometry calculation is closing velocity: the rate at which the gap between the interceptor and the target is shrinking. For a ground-launched kinetic interceptor, this is not just your round's muzzle velocity. It's a vector sum that depends on three factors: the round's own speed, the target's speed along the line of sight, and the geometry of the intercept triangle.
Consider a threat UAS flying a direct approach at 25 m/s at 200 meters range. A kinetic round at 180 m/s muzzle velocity launched directly at the inbound bearing gives you a closing velocity in the range of 195–200 m/s, and time of flight to that range is roughly 1 second. That sounds fast. But at 25 m/s, the drone has moved 25 meters laterally if you've pointed your launcher purely at the detected position rather than the predicted intercept point. A UAS body rarely spans more than 1.5 meters. At engagement ranges of 50–300 meters, the miss distance from failing to lead the target is typically larger than the target itself.
This is not a new insight — it's the same proportional navigation problem that governs missile guidance, scaled down to small-caliber kinetic rounds and shorter time-of-flight regimes. What changes at the counter-UAS scale is the update rate requirement. You can't afford to run guidance corrections over a 3-second flight time like a surface-to-air missile. The time of flight for a short-range kinetic intercept is often under 400 milliseconds. Your lead angle has to be right at launch, which means it has to be computed from a track that is already propagated forward in time by at least that much.
The Lead Angle Computation
Lead angle — the angular offset between the current target bearing and the point in space you aim at — is a function of the ratio of target velocity to interceptor velocity, modulated by the geometry of the intercept. For a target crossing at 90 degrees to the line of sight, the lead angle is approximately arcsin(V_target / V_round). At V_target = 25 m/s and V_round = 200 m/s, that's around 7.2 degrees.
Seven degrees sounds small. But at 200 meters range, 7 degrees of angular error translates to roughly 24 meters of miss distance — which, against a 0.6-meter-wingspan quadrotor, is a complete miss. Get the geometry wrong by 0.5 degrees in your lead angle estimate and you've introduced 1.7 meters of error at 200 meters. Again, larger than the target.
The practical consequence: lead angle computation must be based on a filtered track with velocity confidence, not a raw sensor detection. Two radar returns and a Kalman filter initialization won't give you the velocity estimate precision needed. You need a track history that is long enough to characterize velocity and acceleration, but short enough that the track itself hasn't drifted behind reality. We target a minimum 0.8-second track history before commit, which at a 20 Hz sensor update rate gives roughly 16 position fixes to build velocity and acceleration estimates from.
Engagement Altitude and the Geometry of Approach Angles
Altitude matters in ways that aren't always obvious. A drone at 150 meters altitude on a direct inbound approach at 3 km/h descent creates a substantially different intercept geometry than the same drone at 30 meters altitude. The slant range at first detection changes, the time of flight at equivalent horizontal range changes, and the lead angle calculation shifts because the angular rate of the target across the sensor field of view is now partially in elevation rather than azimuth.
For ground-launched kinetic systems, there's a muzzle elevation constraint — no system fires straight up without serious mechanical complexity. For the kind of fixed-site intercept geometry we're working with, effective altitude engagement bands sit between roughly 15 meters (minimum safe engagement — you don't want intercept debris falling directly onto the protected asset) and 500 meters (where muzzle velocity has decayed enough to affect hit probability and where track quality at that slant range puts lead angle estimates in question).
At very low altitudes — below 30 meters — a direct-inbound threat at close range creates what we call the "engagement dead cone": a zone inside the inner engagement boundary where time of flight is so short that even a correctly computed lead angle cannot be mechanically executed before the drone crosses it. This is not a solvable software problem. It's geometry. The answer is system placement and layered defense, not faster code.
Proportional Navigation at Small Scales
Classical proportional navigation (PN) guidance commands an acceleration proportional to the line-of-sight rotation rate between the missile and the target. PN works well for guided munitions because the missile can make corrections during flight. For unguided kinetic rounds — which is the practical reality for high-volume, low-cost counter-swarm intercept — you get one shot at the launch geometry. There is no in-flight correction.
What we use, conceptually, is a predictive intercept point (PIP) computation at the moment of engagement commit. The PIP is calculated from the current track state (position, velocity, estimated acceleration) propagated forward by the predicted time of flight at the current slant range. The launcher then points at the PIP, not at the current target position. This is functionally equivalent to the lead-pursuit component of PN guidance, applied at launch rather than continuously.
The failure mode of PIP-based intercept is target maneuver between PIP computation and round arrival. Against a cooperative or programmable threat, this is a real concern: a drone executing a simple jink after detecting the launch signature can invalidate a PIP computed even 0.3 seconds earlier. This is why we're skeptical of claims that unguided kinetic intercept alone is sufficient against sophisticated, maneuvering threats. It works well against non-maneuvering and minimally-maneuvering profiles — which covers the majority of the current threat population — but we're not going to claim it solves the entire problem space.
Track Quality as a First-Order Constraint
Everything described above is contingent on track quality. A high-confidence track — one where velocity and acceleration estimates have converged — produces a PIP computation that has a realistic chance of placing a kinetic round within lethal radius of the target. A noisy, initializing, or intermittent track produces a PIP that may be off by meters.
In practice, track quality is degraded by multipath interference at low altitudes (ground clutter), by target aspect changes during terminal approach (affecting radar return strength), and by sensor handoff latency when the track transitions from detection sensor to fire-control sensor. The last one is particularly painful: a track that was being maintained by a wide-area search radar at 10 Hz has to be picked up by a fire-control-quality sensor at 20+ Hz without losing continuity. Any gap in the track during that handoff resets the velocity estimate confidence and may require re-initialization.
Our current approach keeps the track quality metric as a hard gate on engagement commit. If estimated velocity confidence (derived from track covariance) is below threshold, we do not commit to engagement regardless of how close the threat has gotten. This is operationally conservative — it means some threats may reach inner engagement bounds before being engaged — but it avoids wasting rounds on low-confidence intercept attempts that have poor probability of kill.
What the Physics Requires of the System Architecture
Kinematic constraints propagate directly into system design requirements that have no workarounds:
Sensor update rate: Minimum 20 Hz for the track-quality requirement. Below that, velocity estimates at intercept-relevant timescales are too noisy to build a reliable PIP.
Compute latency: The window between engagement commit decision and launcher aiming completion has to fit within the time during which the PIP is valid. For a 200-meter engagement at 25 m/s threat speed, that window is roughly 150–200 milliseconds. End-to-end latency from "go" signal to launcher on-target has to be well inside that.
Launcher slew rate: For crossing targets at angular rates up to approximately 15 degrees/second (a 25 m/s drone at 100 meters is rotating at about 14 degrees/second across the field), the launcher has to be able to track and settle within the required aiming tolerance in under the available slew time. This drives actuator and mount specifications more than almost anything else in the mechanical design.
Engagement sequencing for swarms: Against multiple simultaneous threats, a single kinetic launcher is limited by its own fire rate and the PIP computation time per target. Prioritization — which threat to engage first — has to account for time-to-impact for each track, not just range. The closest drone may not be the most time-critical one if it's also the slowest.
None of these requirements can be relaxed without accepting a corresponding reduction in probability of kill. The physics don't negotiate. What we can do is build a system that meets all of them cleanly, without leaving margin on the table by being slow or imprecise where it doesn't have to be.
Practical Bounds on Probability of Kill
Pk — probability of kill — for kinetic intercept against small UAS is a function of dispersion (the round's accuracy relative to the predicted PIP), the target's effective kill radius (what size error the warhead or round can tolerate and still achieve a lethal hit), and the probability that the track quality was sufficient to produce a valid PIP in the first place.
Published open-source work on small-caliber kinetic intercept suggests Pk values in the 0.4–0.7 range per shot for non-maneuvering targets at engagement ranges under 300 meters, with the wide range reflecting the significant variation in track quality conditions. Multi-shot engagements improve the cumulative Pk substantially, which is why burst-fire and multi-round intercept strategies are standard in current counter-UAS thinking.
Against swarms, individual-shot Pk matters less than throughput: how many engagements can the system execute in a given time window before threats saturate the engagement timeline. A system that achieves 0.85 Pk on single shots but takes 4 seconds per engagement is worse against a 10-unit swarm than one with 0.65 Pk per shot executing at 1.5 seconds per engagement cycle. This is the design problem we're actually solving.
Kinematics set the floor. Everything above the floor is engineering.