Efficient and Compliant Control Framework for Versatile Human-Humanoid Collaborative Transportation

TL;DR

The paper tackles enabling safe and efficient human–humanoid collaboration for object transport in unstructured environments. It introduces a three‑level framework combining an Interaction Linear Inverted Pendulum (I‑LIP) based MPC planner, an admittance‑based compliance model, and a whole‑body QP controller that accounts for coupled robot–object dynamics, with adaptive stiffness modulation to enforce a target relative distance. Key contributions include the I‑LIP model, the admittance shaping, the object‑aware WBC, a novel efficiency metric for dyadic collaboration, and real‑world validation on the Digit humanoid handling a 15 kg load across translation and turning maneuvers. The results demonstrate that local, hand‑level compliance combined with compliant stepping yields higher efficiency and stability, highlighting a practical path toward versatile humanoid co‑transport in logistics, healthcare, and home environments.

Abstract

We present a control framework that enables humanoid robots to perform collaborative transportation tasks with a human partner. The framework supports both translational and rotational motions, which are fundamental to co-transport scenarios. It comprises three components: a high-level planner, a low-level controller, and a stiffness modulation mechanism. At the planning level, we introduce the Interaction Linear Inverted Pendulum (I-LIP), which, combined with an admittance model and an MPC formulation, generates dynamically feasible footstep plans. These are executed by a QP-based whole-body controller that accounts for the coupled humanoid-object dynamics. Stiffness modulation regulates robot-object interaction, ensuring convergence to the desired relative configuration defined by the distance between the object and the robot's center of mass. We validate the effectiveness of the framework through real-world experiments conducted on the Digit humanoid platform. To quantify collaboration quality, we propose an efficiency metric that captures both task performance and inter-agent coordination. We show that this metric highlights the role of compliance in collaborative tasks and offers insights into desirable trajectory characteristics across both high- and low-level control layers. Finally, we showcase experimental results on collaborative behaviors, including translation, turning, and combined motions such as semi circular trajectories, representative of naturally occurring co-transportation tasks.

Figures (9)

• Figure 1: Proposed control framework (right) for collaborative transportation tasks between a human and a humanoid robot (left).
• Figure 2: (a) I-LIP dynamics during the $k^\mathrm{th}$ walking step with the left leg as the stance foot, denoted by $\mathbf{u}_k$. The red and blue solid lines represent spring-damper elements aligned along the local $X'$ and $Y'$ axes, respectively. The red dashed trajectory indicates the evolution of the robot's center of mass (CoM), while the purple dashed trajectory corresponds to the object's CoM. Temporal progression is illustrated through decreasing transparency, where increased opacity represents forward movement in time. The red rectangle denotes the stance foot position, and the yellow rectangle indicates the admissible region for the next foot placement based on kinematic feasibility. (b) Admittance-based interaction model, abstracted as two point masses, representing the CoMs of the robot and the object interconnected by spring damper pairs in the $X'$ and $Y'$ directions. This model captures the compliant coupling between the robot and the object and enables responsive interaction. (c) Top-down view of the I-LIP configuration at the beginning of the $k^\mathrm{th}$ step, highlighting the orientation of spring-damper elements in the horizontal $X'$-$Y'$ plane. (d) Admissible foot placement region with the right leg as the stance foot. The yellow region defines the kinematically feasible area for placing the left foot in the next step. This region is constrained by the robot's reachability and ensures both feasibility and stability in step planning.
• Figure 3: Configuration space of the coupled sytem comprising of Digit and the box.
• Figure 4: Physical box used for the collaboration task. The box itself is made of aluminum and weighs around 15 kg. The hands of Digit are connected to the box through a 3D-printed gauntlet and stainless steel ball joint assembly. The interaction ports at the human and robot side also consist of force torque sensors.
• Figure 5: Results on collaborative in-place walking task. Top: snapshots show the humanoid sharing the load while stepping in place. Bottom: the subscript l in the variables indicates representations in the local frame about the stance foot ($X'$, $Y'$). (a) The distance between the CoMs of the humanoid and the object ($x^\mathrm{b}_l - x^\mathrm{c}_l$) converges to the desired value of 0.6 m. (b) Vertical direction forces applied by the human and the robot ($F^z_\mathrm{h}, F^z_\mathrm{r}$) on the object. (c,d) Modified capture point offsets in the locally forward ($\epsilon^\mathrm{c}_{x,l}$) and lateral direction ($\epsilon^\mathrm{c}_{y,l}$) remain bounded, indicating stability.