A General Purpose Method for Robotic Interception of Non-Cooperative Dynamic Targets

TL;DR

The paper tackles autonomous interception of dynamic, non-cooperative targets using only monocular vision without global localization. It introduces a modular framework combining an EKF-based relative pose estimator, a history-driven target motion predictor, and a receding-horizon trajectory planner that operates in the observer's body frame. The approach demonstrates centimeter-level interception accuracy and high success rates across heterogeneous platforms (rover, UAV, and spacecraft simulators) under partial observability and sensor dropouts, with real-time performance on embedded hardware. It further demonstrates the value of motion prediction in maintaining interception during occlusions and field-of-view losses, and discusses extensions to diffusion-based intent prediction and obstacle handling for future work.

Abstract

This paper presents a general purpose framework for autonomous, vision-based interception of dynamic, non-cooperative targets, validated across three distinct mobility platforms: an unmanned aerial vehicle (UAV), a four-wheeled ground rover, and an air-thruster spacecraft testbed. The approach relies solely on a monocular camera with fiducials for target tracking and operates entirely in the local observer frame without the need for global information. The core contribution of this work is a streamlined and general approach to autonomous interception that can be adapted across robots with varying dynamics, as well as our comprehensive study of the robot interception problem across heterogenous mobility systems under limited observability and no global localization. Our method integrates (1) an Extended Kalman Filter for relative pose estimation amid intermittent measurements, (2) a history-conditioned motion predictor for dynamic target trajectory propagation, and (3) a receding-horizon planner solving a constrained convex program in real time to ensure time-efficient and kinematically feasible interception paths. Our operating regime assumes that observability is restricted by partial fields of view, sensor dropouts, and target occlusions. Experiments are performed in these conditions and include autonomous UAV landing on dynamic targets, rover rendezvous and leader-follower tasks, and spacecraft proximity operations. Results from simulated and physical experiments demonstrate robust performance with low interception errors (both during station-keeping and upon scenario completion), high success rates under deterministic and stochastic target motion profiles, and real-time execution on embedded processors such as the Jetson Orin, VOXL2, and Raspberry Pi 5. These results highlight the framework's generalizability, robustness, and computational efficiency.

Figures (11)

• Figure 1: Examples of real-world (left) and simulated (right) scenarios that our framework can handle on three disparate robotic platforms: a rover ((a) and (b)), a UAV ((c) and (d)), and a spacecraft testbed ((e) and (f)).
• Figure 2: The predicted target trajectory is an element-wise cubic polynomial fit to the most recent EKF pose estimates. We use the EKF estimates as opposed to raw measurements to reduce sensitivity to sensor dropout and noise.
• Figure 3: The orange cone denotes the total reachable area for the observer in a given amount of time, while the blue curve denotes the target's predicted motion. The earliest point at which the blue curve pierces the orange cone is the earliest point at which we can intercept and becomes the goal of our planner. Note that the cross-section of the cone depends on the platform, and in the case of the UAV, is actually a 3D cross-section owing to the additional translational degree of freedom.
• Figure 4: (a) For the rover, the reachable cross-section is peanut-shaped, with the narrow regions corresponding to states that require substantial steering to reach. (b) For the spacecraft testbed, the cross-section is circular due to its holonomic motion; this case is studied in Figure \ref{['fig:reachability']}. (c) For the UAV, the cross-section is spherical, reflecting the additional vertical degree of freedom. For all three cases, the black arrow indicates the observer's current heading.
• Figure 5: An example of the lateral interception case in JPL's Mars Yard. The blue rover is the observer and the red one is the target.