Navigation

• Project Descrption
• Images
• Videos
• Related Publications
• Related Links

Project Description

As part of the Agile Robotics for Logistics (ARL) program, we are developing a planning and control framework for advanced, multi-platform autonomous vehicle operations in an environment of unprecedented complexity. The ARL program seeks to develop and demonstrate semi-autonomous robotics capabilities in an unstructured, outdoor warehouse environment, including cluttered spaces, dynamic obstacles (both humans and other vehicles), and uncertain terrain. The primary platform for this research is a full-scale autonomous forklift, operating within this outdoor warehouse scenario. This autonomous forklift must be able to manipulate and transport pallet loads within this challenging environment without dependence on existing infrastructure, including prior maps or reliable GPS data.

In year 1 of the project, we implemented a hierarchical planning and control strategy to achieve the autonomy necessary to operate within this environment. Using closed-loop rapidly-exploring random trees (RRT), the navigation planner can identify and robustly track obstacle-free trajectories. The planner guarantees waypoint arrival within desired position and heading tolerances, at which point there is a handoff to the manipulation phase. Using a steering controller coupled with perception filters, the manipulation phase guides the forklift to autonomously pick up and drop off arbitrarily placed-pallets, whether on a truck bed or the ground. At a June 2009 demonstration of the prototype forklift at Fort Belvoir, VA, our framework demonstrated accurate path planning capabilities in a realistic warehouse environment.

In year 2 of the project, we extended the forklift's planning and control framework to incorporate higher-level task reasoning and robust navigation. This work leveraged recently proven robustness bounds for the CL-RRT path planning algorithm, ensuring that the vehicle remains safe even when subject to rough terrain and unmodelled dynamics. Modifications to the planner's cost evaluation and perception algorithms yielded intuitive and repeatable behaviors for the vehicle, with minimal risk of collision at all times. Colloborating with other team members, we also significantly expanded the higher-level planning capabilities of the vehicle. The forklift can now queue up multiple tasks in sequence (task planning), memorize the location of previously-observed objects (reacquisition), then retrieve and deliver those objects upon spoken command (spoken vocabulary). In a presentation at Fort Lee, VA in June 2010, the forklift demonstrated reliable and robust operation in an actual supply depot environment, where clutter and tight spaces were prevalent.

In addition to the robotics capabilities demonstrated by the forklift, the ARL program has also developed several support robotics platforms to assist with supply operations. One of these platforms is a small rover, designed and implemented to support the forklift by performing simpler, long-duration tasks, such as a human-guided tour of the warehouse or autonomous inventory checking. This work has expanded the focus of the ARL program to developing multi-robot capabilities in these complex environments, allowing the robots to complete a broader set of tasks with greater efficiency.