💻 computer science

A Unified Complementarity-based Approach for Rigid-Body Manipulation and Motion Prediction

This paper introduces Unicomp, a unified discrete-time modeling framework that integrates free-space motion and complex frictional contact into a single complementarity-based formalism with an ellipsoidal limit surface, enabling stable, real-time optimization-based planning for contact-rich robotic manipulation tasks.

Original authors: Bingkun Huang, Xin Ma, Nilanjan Chakraborty, Riddhiman Laha

Published 2026-02-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Bingkun Huang, Xin Ma, Nilanjan Chakraborty, Riddhiman Laha

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to push a heavy, oddly shaped box across a cluttered floor. Sometimes the box slides smoothly; sometimes it gets stuck because of friction; sometimes it tips over; and sometimes it lifts off the ground entirely.

For a robot to do this, its "brain" (the planning software) usually has to guess what will happen next. Traditional robot brains are like clumsy accountants: they try to separate the world into two neat piles. One pile is "free space" (where the robot moves without touching anything), and the other is "contact" (where things touch and rub). They also often pretend that objects only touch at a single, tiny point, like a needle tip, even when a whole flat side of a box is resting on the floor. This makes the math easy, but the robot's predictions are often wrong, leading to jerky movements or crashes.

The Paper's Big Idea: "Unicomp"
This paper introduces a new software framework called Unicomp. Think of Unicomp as a universal translator that speaks one single language for both free movement and complex touching. Instead of switching between two different rulebooks, it uses one consistent set of mathematical rules (called "complementarity") to handle everything at once.

Here is how it works, using simple analogies:

1. The "Equivalent Contact Point" (The Magic Pivot)

When you push a table with four legs, the contact isn't just one point; it's four. Traditional math tries to track all four, which is messy.

The Paper's Trick: Unicomp pretends all those scattered contact points (like the four legs of a table) are actually just one single "magic pivot point" (called the Equivalent Contact Point or ECP).
The Analogy: Imagine a spinning top. Even though the base is a circle, you can describe its balance by looking at just one central point. Unicomp finds this "center of pressure" for any shape, whether it's a flat box or a wobbly table, and does all the math based on that single point. This lets the robot understand complex shapes without getting confused.

2. The "Friction Ellipsoid" (The Bubble of Possibility)

Friction is tricky. If you push a box, it resists. But if you twist it, it resists differently.

The Paper's Trick: Instead of using a simple "cone" to represent friction (which is a rough guess), Unicomp uses an ellipsoid (a squashed ball shape).
The Analogy: Imagine the friction forces are trapped inside a flexible, squishy balloon.
- If the robot pushes gently, the force stays in the middle of the balloon (the box sticks).
- If the robot pushes hard enough to make the box slide, the force hits the edge of the balloon.
- Crucially, this balloon shape understands that twisting and sliding are connected. It knows that if you twist a box, it might be harder to slide it forward. This captures the "coupled" nature of real-world friction without needing to know exactly how the weight is distributed inside the box.

3. The "Traffic Cop" (Handling Mode Switching)

The hardest part for robots is knowing when to switch from "sticking" to "sliding" or "lifting off."

The Paper's Trick: Unicomp uses a mathematical "switch" that automatically flips based on physics, not human guessing.
The Analogy: Think of a traffic light that changes automatically based on how many cars are waiting. The robot doesn't need to be told, "Okay, now it's time to slide." The math naturally says, "The force is too high for sticking, so we must slide," or "The box is lifting, so friction is gone." It happens instantly and smoothly, allowing the robot to react in real-time (1,000 times a second).

What They Tested

The authors tested this system in a few scenarios:

Pushing weird shapes: They pushed objects with disconnected legs (like a table) and dumbbells. The system handled the "wobbly" support points without crashing.
Speed: It runs fast enough to be used for real-time planning (interactive speeds).
Comparison: They compared it to a popular, standard simulation tool (MuJoCo). They found that while the standard tool sometimes got "jittery" or created fake energy (like a box suddenly jumping up because of a math error), Unicomp stayed smooth and physically realistic.

The Bottom Line

This paper presents a way to teach robots to understand the messy, sticky, complex reality of pushing and pulling objects. By unifying the math for "moving in air" and "rubbing on the floor" into one system, and by using a "magic pivot point" and a "friction balloon" to simplify the geometry, they created a tool that helps robots move more naturally, safely, and quickly in unstructured environments.

Technical Summary: A Unified Complementarity-based Approach for Rigid-Body Manipulation and Motion Prediction (Unicomp)

Problem Statement
Current robotic manipulation frameworks in unstructured environments face significant limitations when reasoning jointly about free-space motion and sustained, frictional contact. Existing local planning and simulation methods typically decouple these regimes or rely on simplified contact representations, such as point contacts or approximated planar patches (e.g., three-point contact). Furthermore, these approaches often approximate friction cones to convert contact problems into Linear Complementarity Problems (LCPs) and frequently exclude manipulator motion planning and obstacle avoidance from the contact dynamics formulation. Consequently, there is a lack of a unified formalism that allows for collision-free trajectory generation at the kinematic level for the robot while simultaneously computing dynamic trajectories for objects at the dynamic level, all while rigorously modeling non-point contact and avoiding friction cone approximations.

Methodology
The paper proposes Unicomp, a unified discrete-time modeling framework that captures both free motion and frictional contact within a single mathematical formalism based on complementarity constraints. The methodology integrates the following components:

Unified Complementarity Framework: The approach formulates free-space motion and contact interactions as coupled Linear and Nonlinear Complementarity Problems (LCPs and NCPs). This allows for principled transitions between contact modes (breaking, sticking, sliding) without enforcing fixed-contact assumptions.
Equivalent Contact Point (ECP) and Ellipsoidal Limit Surface: For planar patch contact, the authors derive a frictional contact model based on the maximum power dissipation principle. Instead of assuming a specific pressure distribution, the set of admissible contact wrenches is represented by an ellipsoidal limit surface. This representation captures coupled force–moment effects, including torsional friction, and is agnostic to the underlying pressure distribution. The contact is modeled via an Equivalent Contact Point (ECP) where the wrench is applied.
Discrete-Time Dynamics: The system employs a forward Euler time-stepping scheme. The Newton-Euler equations are discretized to relate generalized velocities and contact wrenches through quadratic constraints. The contact impulse is parameterized by normal, tangential, and torsional components, solved via a Mixed Complementarity Problem (MCP).
Whole-Body Collision Avoidance: Collision avoidance for the manipulator is cast as a complementarity-based projection on the same discrete-time grid. The robot's geometry is approximated by a set of spheres. Unilateral separation constraints are formulated as a Linear Complementarity Problem (LCP) to correct nominal joint increments, ensuring collision-free trajectories at the kinematic level.
Integration: The framework combines these elements into a modular pipeline (Algorithm 3) that simultaneously solves for object dynamics (via NCP/MCP) and robot kinematics with obstacle avoidance (via LCP).

Key Contributions

Unified Formulation: The paper presents a real-time motion prediction framework that considers both free-space motion with collision avoidance and intermittent contact dynamics within a single complementarity-based structure.
Non-Point Contact Modeling: It rigorously accounts for non-point (planar patch) contact without approximating friction cone constraints, utilizing an ellipsoidal limit surface derived from the maximum power dissipation principle.
Coupled Kinematic-Dynamic Planning: The framework enables the generation of collision-free robot trajectories at the kinematic level while simultaneously computing object trajectories at the dynamic level, bridging a gap in existing software implementations.
Real-Time Execution: The formulation yields a discrete-time predictive model suitable for real-time optimization-based planning, demonstrated at interactive speeds (1000 Hz).

Experimental Results
The authors validate the approach through numerical simulations and comparisons:

Non-Prehensile Interactions: Experiments with direct center-of-mass forcing and tool-driven pushing demonstrate that the formulation captures intermittent contact and stick–slip transitions without prescribing discrete modes. The results show consistent contact signatures where sliding saturates the limit surface ( $s \approx 1$ ) and sticking remains inside ( $s < 1$ ).
Non-Convex Contact: The system successfully handles interactive planar pushing with non-convex objects (e.g., table-like and dumbbell-like shapes) where support regions switch abruptly. The ECP migrates across distinct ground regions while maintaining physically consistent motion at 1000 Hz.
Comparison with MuJoCo: In comparative experiments, Unicomp demonstrated smoother trajectories and greater stability than MuJoCo during sliding and breaking modes. Specifically, MuJoCo exhibited "phantom forces" and non-physical energy injection (rapid kinetic energy increase) during ground penetration events, whereas Unicomp remained stable.
Multi-Body Propagation: The framework successfully models impulse propagation in multi-body scenes (tool $\to$ block 1 $\to$ block 2) without explicit enumeration of contact modes.
Planning Applications: The authors implemented a closed-loop planar pushing planner and a whole-body obstacle-aware pushing system, demonstrating the ability to drive objects to desired poses while avoiding collisions.

Significance and Claims
The paper claims that Unicomp provides a practical foundation for robust manipulation planning in unstructured environments. By unifying free-space motion and contact-rich behaviors under a single complementarity-based formalism, the approach enables stable, physically consistent behavior at interactive speeds. The authors emphasize that their method avoids the fidelity limitations of simplified contact representations and the decoupling of kinematic and dynamic planning, thereby supporting the robust execution of complex, contact-rich behaviors in real time. The work is presented as a constructive evidence that such a formulation is feasible for real-time execution of complex robotic maneuvers.