Think Fast: Real-Time Kinodynamic Belief-Space Planning for Projectile Interception

TL;DR

The paper tackles real-time interception of fast-moving projectiles under sensor noise by introducing a tree-like kinodynamic planner built from minimum-time, jerk-bounded motion primitives in state-time space, coupled with an Innovation-based Adaptive Estimation Adaptive Kalman Filter to maintain a belief over target trajectories.A first-order delta method yields a Gaussian approximation of the crossing time and intercept plane state, enabling online belief updates over a reduced goal space while the robot acts.An offline action-tree synthesis and a QMDP-based value function allow fast online decision-making (sub-10 ms) that can adapt as observations evolve, demonstrated on a 6-DOF ABB arm with a shield and a ZED 2i camera.Hardware experiments show the method improves interception success over a naive approach (74% vs 63%), illustrating the value of uncertainty-aware planning under tight temporal constraints, though future work aims to increase manipulability, move toward continuous action spaces, and support multiple interceptors.

Abstract

Intercepting fast moving objects, by its very nature, is challenging because of its tight time constraints. This problem becomes further complicated in the presence of sensor noise because noisy sensors provide, at best, incomplete information, which results in a distribution over target states to be intercepted. Since time is of the essence, to hit the target, the planner must begin directing the interceptor, in this case a robot arm, while still receiving information. We introduce an tree-like structure, which is grown using kinodynamic motion primitives in state-time space. This tree-like structure encodes reachability to multiple goals from a single origin, while enabling real-time value updates as the target belief evolves and seamless transitions between goals. We evaluate our framework on an interception task on a 6 DOF industrial arm (ABB IRB-1600) with an onboard stereo camera (ZED 2i). A robust Innovation-based Adaptive Estimation Adaptive Kalman Filter (RIAE-AKF) is used to track the target and perform belief updates.

Figures (7)

• Figure 1: ABB arm with a shield attached to its end-effector blocking a thrown projectile
• Figure 2: Task Space Goal Poses
• Figure 3: Action Tree for 2D Double Integrator
• Figure 4: Flowchart showing Perception and State Estimation
• Figure 5: The left and right figures show 3D locations of the projectile at different throws. Measurement noise varies between throws.