Avadesh Meduri,Paarth Shah,Julian Viereck,Majid Khadiv,Ioannis Havoutis,Ludovic Righetti
Abstract:Online planning of whole-body motions for legged robots is challenging due to the inherent nonlinearity in the robot dynamics. In this work, we propose a nonlinear MPC framework, the BiConMP which can generate whole body trajectories online by efficiently exploiting the structure of the robot dynamics. BiConMP is used to generate various cyclic gaits on a real quadruped robot and its performance is evaluated on different terrain, countering unforeseen pushes and transitioning online between different gaits. Further, the ability of BiConMP to generate non-trivial acyclic whole-body dynamic motions on the robot is presented. The same approach is also used to generate various dynamic motions in MPC on a humanoid robot (Talos) and another quadruped robot (AnYmal) in simulation. Finally, an extensive empirical analysis on the effects of planning horizon and frequency on the nonlinear MPC framework is reported and discussed.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the challenges faced when legged robots perform online whole - body motion planning. Specifically, due to the inherent nonlinearity of robot dynamics, it is very difficult to generate whole - body motions online. For this reason, the paper proposes a nonlinear model predictive control framework - BiConMP (Biconvex Model Predictive Control), which can generate whole - body trajectories online by efficiently utilizing the structure of robot dynamics.
### Main Problems
1. **Challenges in Online Whole - Body Motion Planning**:
- The dynamics of legged robots are nonlinear, which makes the online generation of whole - body motions complex.
- Existing methods usually require pre - defined gait sequences, which limit the flexibility and adaptability of the robot.
- Many existing methods perform poorly when dealing with complex terrains and external disturbances.
2. **Requirement for Real - Time Performance**:
- In order for the robot to make real - time decisions in a changing environment, the algorithm needs to run fast enough.
- Existing nonlinear optimization methods are usually computationally expensive and difficult to be applied in real - time.
### Solutions
- **BiConMP Framework**:
- It takes advantage of the biconvex structure of robot dynamics, decomposes the nonlinear optimization problem into two convex sub - problems, and optimizes the states (such as the position of the center of mass, velocity, angular momentum) and forces (such as contact forces) respectively.
- It uses the Alternating Direction Method of Multipliers (ADMM) to efficiently solve these two convex sub - problems, ensuring the fast convergence and real - time performance of the algorithm.
- It uses a custom - implemented Fast Iterative Shrinkage - Thresholding Algorithm (FISTA) to solve each convex sub - problem, ensuring a quadratic convergence rate and effectively imposing various constraints.
- **Practical Applications**:
- It has achieved the online generation of multiple gaits (such as trotting, jumping, running) at a frequency of 20 Hz on an actual quadruped robot (such as Solo12).
- The framework has demonstrated robustness to different terrains and external disturbances.
- It has also demonstrated applications on a humanoid robot (such as Talos) and another quadruped robot (such as AnYmal) in simulations.
### Summary
The main contribution of this paper is the proposal of an efficient nonlinear model predictive control framework that can achieve real - time whole - body motion planning, which is suitable for the autonomous navigation and operation of legged robots in complex environments. By taking advantage of the biconvex structure of dynamics and advanced optimization algorithms, the BiConMP framework performs excellently in terms of computational efficiency and real - time performance, providing a new solution for the online motion planning of legged robots.