qpOASES is an open-source C++ library for solving Quadratic Programming (QP) problems, designed for embedded applications with millisecond-scale execution times. Created by Hans Joachim Ferreau and prof. Moritz Diehl in the OPTEC group at KU Leuven and KIT (Karlsruhe Institute of Technology), currently hosted by the COIN-OR foundation and maintained by the academic community.
The main advantage of qpOASES is its implementation of the online active-set strategy: after the first QP solution, the solver maintains an active-constraints basis and on the next call (with slightly changed parameters) performs only a small number of pivot steps. This makes qpOASES an ideal solver for MPC (Model Predictive Control), where the same structural QP problem is solved in a loop at 100 Hz - 1 kHz.
The library is free and open-source under LGPL 2.1, written in pure C++ with minimal dependencies (only BLAS/LAPACK). Multi-language support: C, Python (qpoases-python), MATLAB/Simulink, Octave, dSPACE. qpOASES is part of the ACADO Toolkit (for nonlinear MPC) and CasADi (auto-differentiation optimization framework), where it is one of the preferred QP solvers.
In humanoid robotics and autonomous vehicles, qpOASES is one of the dominant solvers for Whole Body Control (WBC) — a control problem formulation that expresses the entire robot's body control as a QP with hierarchical tasks and physical constraints (contacts, torque limits, joint limits). Notable deployments: KIT ARMAR-6 and ARMAR-7, DLR Justin, IHMC Atlas, Boston Dynamics (early research projects), DARPA Robotics Challenge — most teams used qpOASES as their WBC solver.
Unlike more general solvers (CPLEX, Gurobi, OSQP), qpOASES specializes in small-scale problems (up to ~1000 decision variables) with dense matrices — the typical range for robot MPC. For larger sparse problems OSQP is a better choice. qpOASES has slow release cadence (~1-2 years between releases) but is a fundamental tool in embedded optimization and very mature code (TRL 9, stable API since 2014).
An API Library is a software package that exposes programmatic interfaces for communicating with a device, service, or system. In robotics it typically forms a lightweight integration layer built on top of the manufacturer's official API or an open-source project, abstracting low-level protocol details and providing language-native bindings (Python, C++, Java, etc.).
A Control Stack is the set of software components responsible for control logic, motion planning, command execution, and coordination of a robot's actuators. It receives high-level goals or waypoints (from planning or AI layers) and translates them into low-level actuator commands, closed-loop feedback control, and safety checks. Common implementations include MoveIt (arm manipulation), Nav2 (mobile navigation), and custom whole-body control stacks for humanoids.
Robot Control denotes the role of software responsible for motion control, command execution, coordination of actuating elements and the direct operational logic of the robot.
A family of open-source libraries for motion planning (RRT, PRM, Lazy-PRM), inverse kinematics, rigid-body dynamics, and trajectory optimization: MoveIt, OMPL, Pinocchio, Drake, RBDL, TrajOpt, KDL, Bullet, iDynTree.
KIT ARMAR-6, ARMAR-7 (KIT High-Performance Humanoid Technologies Lab, Karlsruhe) — humanoid laboratory robots using qpOASES as their Whole Body Control solver. DLR Justin / Rollin' Justin (German Aerospace Center) — humanoid torso used for space manipulation research, qpOASES as the MPC solver for torso control. IHMC Atlas (DARPA Robotics Challenge 2015) — Florida Institute for Human and Machine Cognition, runner-up in the DRC, qpOASES in the WBC stack. ACADO Toolkit / CasADi — used as the embedded QP solver in both nonlinear MPC optimization frameworks. Mercedes-Benz, Daimler AG — autonomous driving experiments 2014-2018 (Bertha Drive Project), qpOASES in lateral MPC at highway speeds. Numerous academic projects in legged robotics (CMU, ETH Zurich, IIT Genoa, KAIST) and autonomous racing (TUM, Formula Student).
GitHub: coin-or/qpOASES — ~400 stars, ~120 forks, ~25 contributors, sporadic updates, mature codebase. Academic community — several thousand citations in robotics, MPC and autonomous driving papers. The original publication Ferreau, Kirches, Potschka, Bock, Diehl (2014, Math. Prog. Comp.) is cited >2500 times. Maintenance — mostly volunteers from KIT and University of Freiburg (Diehl Lab). COIN-OR mailing list with low activity. No commercial support — purely academic project. Ecosystem — embedded in ACADO, CasADi, where it has tens of thousands of downstream users across robotics, automotive and chemical-process control.
Boston Dynamics bipedal humanoid robot. The fully electric generation unveiled in 2024 succeeds the hydraulic Atlas that was retired after more than a decade of research.
Tutor Intelligence industrial robot designed for palletizing, depalletizing and case-picking in warehouses and factories, offered as Robotics-as-a-Service from $14/hr.
Full-size general-purpose humanoid (~180 cm, ~47 kg). Bipedal, 5 DOF per leg, 3.3 m/s walking speed, 360° perception via 3D LiDAR + depth camera, Unitree M107 PMSM joint motors (~360 N·m peak knee torque).
Ubuntu 24.04 LTS 'Noble Numbat' — supported until April 2029. The host for ROS 2 Jazzy.
Requires BLAS/LAPACK on host (e.g. OpenBLAS, Intel MKL). For embedded use (qpOASES Embedded) a standalone version without external dependencies is available.
License family: Weak Copyleft
MATLAB interface update, embedded variant fixes, improved Windows support.
Support for the OQP (Online QP) variant with dynamic allocation, new Simulink interface.
API stabilization, extended regression tests, sparse representation support.
First COIN-OR version (transition from KIT). Full C++11 support, CMake build system.
Core refactor, new Python interface (qpoases-python), support for multiple constraint matrices.
First public release from KIT — Ferreau et al. online active-set strategy implementation.
