Stochastic integrated task and motion planning for human‑robot collaboration

About the project

STAMP is carried out at the Laboratory for Autonomous Systems and Mobile Robotics (LAMOR), University of Zagreb Faculty of Electrical Engineering and Computing (FER), and runs from 1 October 2026 to 30 September 2029.

The project aims to enable safe and efficient human-robot collaboration (HRC) in dynamic, uncertain environments, which is a fundamental challenge in robotics. Robots perform well in structured industrial settings, but they struggle in unstructured spaces where human presence introduces unpredictability.

STAMP proposes a novel stochastic integrated task and motion planning (TAMP) framework that explicitly models uncertainty throughout the planning hierarchy. Unlike deterministic TAMP approaches, STAMP propagates uncertainty from task-level decisions through motion planning, enabling risk-aware action selection. The robot maintains probabilistic beliefs about human intentions and future states, which leads to anticipatory re-planning when predictions indicate potential conflicts.

The project will use motion capture systems, RGB-D sensors and collaborative robot platforms. All developed methods will be released as open-source, ROS-compatible software.

This work has been supported by the Croatian Science Foundation under the project UIP-2026-8339.

Objectives

The overall objective of STAMP is to develop and validate a stochastic integrated task and motion planning framework that enables robot arms to collaborate with humans safely, efficiently and robustly in dynamic and uncertain environments.

	Objective
O1	Motion planning. Improve efficacy of motion planning approaches for robot arms in HRC scenarios using continuous-time Gaussian-process trajectory optimization with a constant-jerk prior.
O2	Human motion prediction. Exploit contextual information for accurate human motion prediction: intention estimation and full-body trajectory forecasting with calibrated uncertainty.
O3	Stochastic task planning. Develop a stochastic task planning framework for long-term autonomy, using gradient-free optimization guided by deep generative priors.
O4	Robust execution. Optimize low-level execution with model predictive control that accounts for uncertainty and for forces induced by grasped objects.
O5	Real-world demonstrations. Demonstrate the framework in real-world, high degree-of-freedom HRC scenarios: handovers, collaborative box stacking and peg-in-hole insertion.