Bilevel Learning for Dual-Quadruped Collaborative Transportation under Kinematic and Anisotropic Velocity Constraints

TL;DR

This work tackles dual-quadruped collaborative payload transportation under payload-induced kinematic coupling and anisotropic velocity limits. It introduces BLCT, a constrained bilevel optimization framework that jointly learns a team-level collaboration policy (upper level) and individual robot velocity controls (lower level) using PPO, while enforcing payload kinematics, obstacle collisions, and direction-dependent speed limits. The approach demonstrates superior success rates and efficient planning compared to baseline planners and RL, and it validates the method in both Gazebo simulations and real two-quadruped experiments, including tight-space reconfiguration behaviors. The results suggest BLCT's potential to enable reliable, efficient multi-robot collaboration for complex, constrained transportation tasks with practical applicability in warehouse and industrial settings.

Abstract

Multi-robot collaborative transportation is a critical capability that has attracted significant attention over recent years. To reliably transport a kinematically constrained payload, a team of robots must closely collaborate and coordinate their individual velocities to achieve the desired payload motion. For quadruped robots, a key challenge is caused by their anisotropic velocity limits, where forward and backward movement is faster and more stable than lateral motion. In order to enable dual-quadruped collaborative transportation and address the above challenges, we propose a novel Bilevel Learning for Collaborative Transportation (BLCT) approach. In the upper-level, BLCT learns a team collaboration policy for the two quadruped robots to move the payload to the goal position, while accounting for the kinematic constraints imposed by their connection to the payload. In the lower-level, BLCT optimizes velocity controls of each individual robot to closely follow the collaboration policy while satisfying the anisotropic velocity constraints and avoiding obstacles. Experiments demonstrate that our BLCT approach well enables collaborative transportation in challenging scenarios and outperforms baseline approaches.

Figures (5)

• Figure 1: A motivating scenario for a pair of quadruped robots collaboratively transport an oversized payload in warehouse environments. The robots must not only make team decisions for collaboration but also generate individual navigational controls, while satisfying multiple constraints due to quadruped and payload kinematics, boundary collisions, and anisotropic velocity limits.
• Figure 2: Overview of our BLCT method. By formulating collaborative transportation as a bilevel learning problem, BLCT uses an upper-level optimization to learn collaborations and a lower-level optimization to generate robot controls, while considering the kinematic, collision, and anisotropic velocity constraints.
• Figure 3: Overall performance of BLCT and comparisons with the A*, RRT*, and RL methods. BLCT meets or outperforms baseline performance across the four evaluation metrics.
• Figure 4: Qualitative results obtained by BLCT running on two quadruped robots in Gazebo simulation when the robots collaboratively transport a payload in the right turn scenario. Additional qualitative results for the forward bottleneck and left turn scenarios are on the project website.
• Figure 5: Qualitative results obtained by BLCT deployed on two physical Unitree Go1 robots when the robots collaboratively transport a wooden payload in the right turn scenario. Additional qualitative results for the forward bottleneck and left turn scenarios are on the project website.