Touch-Based Object Localisation with Spatially-Aware Belief Entropy Estimation
Lara Brudermüller, Julius Jankowski, Marc Toussaint, Nick Hawes
AI summary
Problem
Vision-based localisation fails in occluded or cluttered environments, while existing touch-based methods struggle with high-dimensional continuous state spaces, sparse binary contact signals, and particle starvation.
Approach
The method employs a continuous-space particle filter with a proximity-aware measurement model and contact-aware resampling, guided by an information-gathering controller that maximises expected information gain using a non-parametric entropy estimator.
Key results
- Continuous-space particle filter with proximity-aware measurement model
- Contact-aware resampling mitigates particle starvation under sparse binary contacts
- Non-parametric entropy estimator captures both observation and dynamics-driven belief changes
- Reliable localisation and grasping on real hardware from initial beliefs with up to 0.4 m mode separation
Why it matters
Enables robust robotic manipulation in visually challenging or vision-denied environments by scaling touch-based localisation beyond narrow uncertainty assumptions.
Abstract
Robust robotic manipulation in the real world requires coping with incomplete or unreliable sensory in- put. While vision provides rich information, it often fails in the presence of occlusions, clutter, or poor lighting. In such cases, touch offers a robust alternative, enabling object localisation through contact alone. We present a touch-only global localisation method that operates in continuous state space with a particle belief. Sparse contact/no-contact signals are turned into informative likelihoods via a proximity-aware measurement model, and contact-aware resampling mitigates particle starvation. An information-gathering controller selects actions that maximise expected information gain using a non- parametric entropy estimator sensitive to both observation updates and dynamics. On real hardware, the system reliably localises and then grasps from broad, multi-modal initial beliefs with mode separations up to 0.4 m, far beyond the narrow uncertainty ranges assumed in related work. Information-aware localisation-actions speed up belief convergence and boost grasp success; and ablations in simulation confirm the benefits of the measurement and resampling components.