Research Analyzer
← Back ICRA 2026

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

Gabriel Olin, Lu Chen, Nayesha Gandotra, Maxim Likhachev, Howie Choset

PDF

AI summary

Key figure (auto-extracted from paper)
Explicitly modeling sensor uncertainty in real-time belief-space planning significantly improves robotic projectile interception success rates under tight time constraints.
belief-space planning kinodynamic planning real-time interception uncertainty reasoning robotic catching motion primitives

Problem

Intercepting fast-moving objects requires split-second decisions under noisy, incomplete sensor data, yet existing planners rarely reason about uncertainty in real time, leading to poor success rates.

Approach

The authors build a tree-based kinodynamic planner using minimum-time motion primitives in state-time space, coupled with an adaptive Kalman filter to continuously update a Gaussian belief over the target's trajectory and dynamically select optimal intercept goals.

Key results

  • 74% interception success rate on hardware versus 63% for a naive deterministic baseline
  • Online planning computations completed in under 10 milliseconds per step
  • Demonstrated robust real-time interception on a 6-DOF ABB IRB-1600 robot arm with a stereo camera
  • Validated that hedging against belief distributions outperforms committing to the first observed trajectory

Why it matters

Demonstrates that real-time uncertainty reasoning is critical for reliable robotic interception, offering a deployable framework for time-critical manipulation tasks.

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.

Index terms

Reactive and Sensor-Based Planning Planning under Uncertainty Motion and Path Planning

Related papers