Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Calibrated Particle-sets for Trans-dimensional Uncertainty Representation (CaPTURe), a novel algorithm that enhances uncertainty calibration in robotic systems interacting with complex environments. By leveraging a calibration dataset of system transitions, CaPTURe effectively accounts for the multimodal nature of future configurations, particularly in contact-rich scenarios, ensuring that probabilistic prediction regions are both accurate and reliable. The method demonstrates a significant improvement in task success rates, achieving up to a 30% absolute increase over existing baselines in simulated planning tasks involving obstacle navigation and precision manipulation.
CaPTURe achieves up to a 30% improvement in task success rates by accurately calibrating uncertainty in contact-rich environments, a critical factor for safe robotic interactions.
Reliable uncertainty representation is essential for deploying autonomous systems that interact with their environment, as robots must reason about how uncertainty arising from both stochasticity and model mismatch is impacted by contacts with obstacles (e.g., when navigating through a cluttered environment or inserting a part into an assembly). We propose Calibrated Particle-sets for Trans-dimensional Uncertainty Representation (CaPTURe), a geometry-aware, conformal prediction-based algorithm that generates probabilistically valid prediction regions of the unknown future system configuration using particle-based models of arbitrary fidelity. While calibrated uncertainty predictions are essential for safe and efficient planning, analytical or learned motion models are often inaccurate - due to limited data, simplifying assumptions, unmodeled effects, etc. - which can lead to unsafe executions or task failure. Additionally, when a robot contacts an obstacle, the distribution of its future configurations can become multimodal or disjoint, or lie along manifolds of lower intrinsic dimension than the space of possible robot configurations. Our method uses a calibration dataset of system transitions to locally calibrate motion uncertainty estimates, constructing regions guaranteed to contain the future robot configuration at a user-set probability. Our calibration procedure captures how motion uncertainty varies between contact-rich and contactless motions, leading to sufficient coverage in both cases. We evaluate our method on two simulated planning tasks: controlling a marble around a labyrinth and performing tight-tolerance peg-in-hole insertion with a manipulator. Compared to relevant baselines, CaPTURe achieves the user-specified coverage requirement both in and out of contact and achieves up to a 30% absolute improvement in task success rate over the best baseline.