Dynamic Walking with Compliance on a Cassie Bipedal Robot (1904.11104v1)

Abstract: The control of bipedal robotic walking remains a challenging problem in the domains of computation and experiment, due to the multi-body dynamics and various sources of uncertainty. In recent years, there has been a rising trend towards model reduction and the design of intuitive controllers to overcome the gap between assumed model and reality. Despite its viability in practical implementation, this local representation of true dynamics naturally indicate limited scalibility towards more dynamical behaviors. With the goal of moving towards increasingly dynamic behaviors, we leverage the detailed full body dynamics to generate controllers for the robotic system which utilizes compliant elements in the passive dynamics. In this process, we present a feasible computation method that yields walking trajectories for a highly complex robotic system. Direct implementation of these results on physical hardware is also performed with minimal tuning and heuristics. We validate the suggested method by applying a consistent control scheme across simulation, optimization and experiment, the result is that the bipedal robot Cassie walks over a variety of indoor and outdoor terrains reliably.

• The paper introduces an advanced full-body dynamic model incorporating compliant elements for improved control of the Cassie robot.
• The paper employs nonlinear programming using FROST to optimize robust trajectories within physical constraints.
• The paper validates the approach experimentally, confirming reliable robot walking across diverse terrains with precise metrics.

Dynamic Walking with Compliance on a Cassie Bipedal Robot

The paper, "Dynamic Walking with Compliance on a Cassie Bipedal Robot," authored by Jacob Reher, Wen-Loong Ma, and Aaron D. Ames, explores an innovative approach to controlling bipedal robots, specifically focusing on the Cassie robot. The core challenge addressed by the paper is the discrepancies between simplified dynamic models and the complex physical systems they aim to replicate. The authors propose utilizing a detailed full-body dynamic model to create more accurate control schemes, which incorporate compliant elements intrinsic to the robot's passive dynamics. This strategy is contrasted against conventional methods that generally simplify these dynamics.

Key Contributions

1. Comprehensive Robot Modeling: The Cassie robot is modeled as a constrained dynamical system embedded with compliance, expanded upon both simple and full models. The simple model treated leaf springs as rigid links, whereas the full model considered them as torsion joints, adding complexity to the geometric and dynamic relations. The authors provide a detailed hybrid dynamic model, segmenting the robot's walking cycle into discrete domains and developing a set of equations to represent the system’s motion.
2. Trajectory Optimization: The authors use a nonlinear programming approach to solve for optimal walking trajectories. They leverage an optimization framework (FROST) to compute feasible motion paths for these complex dynamics, considering physical limitations such as torque bounds and friction constraints. This optimization process results in a consistent and realizable motion trajectory that manifests in reduced experimental tuning requirements.
3. Implementation and Experiments: The paper outlines practical implementation on hardware via a dual-threaded control and estimation framework, enabling efficient computation on real-time systems. The authors validate their approach experimentally by demonstrating Cassie's ability to walk reliably in various environments. The results, meticulously presented, align closely with the formulated computational predictions.

Numerical and Experimental Insights

The numerical optimization results revealed that the full model, despite being computationally intensive, is beneficial for capturing the finer details of robotic compliance, manifesting more accurate and robust control schemes. Experiments exhibited Cassie's successful navigation across diverse terrains, validating the effectiveness of the compliant model. Key metrics included the measurement of spring deflections, limit cycles, and kinetics of various joint movements under real-world conditions, which consistently matched the optimization expectations.

Implications and Future Work

This research underscores the significance of deploying detailed dynamic models when designing controllers for complex robotic systems, particularly in environments where compliance plays a crucial role. The implications are promising for robotic applications, suggesting enhanced reliability in natural and unpredictable environments, such as off-road navigation. The paper sets the stage for future advancements in refining computational efficiency and adaptability of robot control systems in dynamic settings. The paper suggests that ongoing computational strategies must address the balance between model complexity and experimental convenience, paving the way for further advances in robust and adaptive bipedal locomotion strategies.