Safe Planner: Empowering Safety Awareness in Large Pre-Trained Models for Robot Task Planning

TL;DR

A novel Safe Planner framework is proposed, which empowers safety awareness in large pre-trained models to accomplish safe and executable planning and achieves state-of-the-art task success rates, but also substantially improves safety during task execution.

Abstract

Robot task planning is an important problem for autonomous robots in long-horizon challenging tasks. As large pre-trained models have demonstrated superior planning ability, recent research investigates utilizing large models to achieve autonomous planning for robots in diverse tasks. However, since the large models are pre-trained with Internet data and lack the knowledge of real task scenes, large models as planners may make unsafe decisions that hurt the robots and the surrounding environments. To solve this challenge, we propose a novel Safe Planner framework, which empowers safety awareness in large pre-trained models to accomplish safe and executable planning. In this framework, we develop a safety prediction module to guide the high-level large model planner, and this safety module trained in a simulator can be effectively transferred to real-world tasks. The proposed Safe Planner framework is evaluated on both simulated environments and real robots. The experiment results demonstrate that Safe Planner not only achieves state-of-the-art task success rates, but also substantially improves safety during task execution. The experiment videos are shown in https://sites.google.com/view/safeplanner .

Figures (6)

• Figure 1: An illustrative example of safe planning.
• Figure 2: The Safe Planner framework. First, our framework translates the natural language task instruction into the PDDL goal predicates. Then, the goal predicates are combined with the PDDL domain as the prompt for the VLM planner. Note that in the Safe Planner framework, the VLM planner not only takes the observations as inputs, but also considers the safety of the operators, and selects the next operator to execute.
• Figure 3: The robot and target task scenes in the simulated experiments.
• Figure 4: The task scene for real robot experiments.
• Figure 5: Comparison of the skill sequences planned by the model with SM (left) and without SM (right).