Linzhu Yue,Zhitao Song,Jinhu Dong,Zhongyu Li,Hongbo Zhang,Lingwei Zhang,Xuanqi Zeng,Koushil Sreenath,Yun-hui Liu
Abstract:Natural terrain complexity often necessitates agile movements like jumping in animals to improve traversal efficiency. To enable similar capabilities in quadruped robots, complex real-time jumping maneuvers are required. Current research does not adequately address the problem of online omnidirectional jumping and neglects the robot's kinodynamic constraints during trajectory generation. This paper proposes a general and complete cascade online optimization framework for omnidirectional jumping for quadruped robots. Our solution systematically encompasses jumping trajectory generation, a trajectory tracking controller, and a landing controller. It also incorporates environmental perception to navigate obstacles that standard locomotion cannot bypass, such as jumping from high platforms. We introduce a novel jumping plane to parameterize omnidirectional jumping motion and formulate a tightly coupled optimization problem accounting for the kinodynamic constraints, simultaneously optimizing CoM trajectory, Ground Reaction Forces (GRFs), and joint states. To meet the online requirements, we propose an accelerated evolutionary algorithm as the trajectory optimizer to address the complexity of kinodynamic constraints. To ensure stability and accuracy in environmental perception post-landing, we introduce a coarse-to-fine relocalization method that combines global Branch and Bound (BnB) search with Maximum a Posteriori (MAP) estimation for precise positioning during navigation and jumping. The proposed framework achieves jump trajectory generation in approximately 0.1 seconds with a warm start and has been successfully validated on two quadruped robots on uneven terrains. Additionally, we extend the framework's versatility to humanoid robots.
What problem does this paper attempt to address?
This paper attempts to solve the problem of real - time omnidirectional jumping of quadruped robots on complex terrains. Specifically, current research has deficiencies in online omnidirectional jumping trajectory planning, ignoring the dynamic and kinematic constraints of robots during the trajectory generation process. In addition, existing methods often rely on offline optimization, cannot meet the requirements of real - time applications, and are usually limited to specific scenarios. Therefore, this paper proposes a general and complete cascaded online optimization framework for the omnidirectional jumping of quadruped robots. This framework systematically covers the design of jumping trajectory generation, trajectory tracking controllers, and landing controllers, and combines environmental perception technology to enable robots to avoid obstacles that cannot be bypassed by standard walking methods, such as jumping from a high platform.
### Main contributions:
1. **Development of omnidirectional jumping framework**:
- A general and comprehensive quadruped omnidirectional jumping trajectory framework is proposed, including online trajectory optimization, real - time trajectory tracking, and active compliant landing controllers. This framework supports multi - directional jumps, including forward, backward, left, right, and diagonal directions (such as forward - right). By integrating environmental perception, this framework enables quadruped robots to overcome obstacles by jumping on uneven terrains. Moreover, this method has also been successfully extended to support bipedal jumping.
2. **Novel online - optimized jumping problem definition and solution**:
- By constructing an omnidirectional jumping plane and using a reduced - order model to define the optimization problem, the ground reaction forces (GRFs) and the center of mass (CoM) trajectories are generated. Dynamic and kinematic constraints are considered during the optimization process to ensure dynamic and kinematic feasibility. To achieve real - time optimization within approximately 0.1 seconds, a differential evolution algorithm is proposed, combined with configuration space analysis, Latin hypercube sampling (LHS) initial sampling, pre - motion libraries as warm - start strategies, and an innovatively designed fitness function.
3. **High - level positioning robustness under large impacts during landing**:
- To ensure the accuracy of environmental - perception - assisted jumping on uneven terrains, a robust positioning recovery mechanism based on a coarse - to - fine re - positioning method is introduced. In the coarse - positioning stage, the global branch - and - bound (BnB) search algorithm is used to provide an accurate initial pose; in the fine - tuning stage, the maximum a posteriori (MAP) estimation is used to fuse IMU, motor encoder, and LiDAR sensor data to achieve precise positioning.
4. **Extensive experimental verification**:
- Through extensive simulations and practical experiments, including multi - directional jumps and autonomous navigation (indoor and outdoor), the robustness and practicality of the algorithm are verified. In addition, the jumping framework has also been extended to humanoid robots in a simulated environment, demonstrating its diversity and effectiveness.
### Key technical details:
- **Omnidirectional jumping model**: The omnidirectional jump is divided into three stages: take - off, flight, and landing. By defining a jumping plane J - Z perpendicular to the X - Y plane and assuming that the force perpendicular to this plane is zero, the complexity of omnidirectional jumping planning is simplified.
- **Trajectory optimization**: The ground reaction forces (GRFs) are optimized by polynomial curves and converted into optimizing the center of mass trajectory, the time of different jumping stages, and the GRFs. Optimization variables include the center of mass position, rotation angle, and the time of each stage.
- **Dynamic and kinematic constraints**: Dynamic and kinematic constraints are considered during the optimization process to ensure that the generated trajectories are physically feasible.
- **Environmental perception and re - positioning**: Combining environmental perception technology enables robots to jump on uneven terrains and improves the accuracy of positioning through a coarse - to - fine re - positioning method.
The framework proposed in this paper not only solves the technical problems of real - time omnidirectional jumping of quadruped robots on complex terrains but also provides new ideas and methods for future research.