An official website of the United States government
Legged robots with point or small feet are nearly impossible to control instantaneously but are controllable over the time scale of one or more steps, also known as step-to-step control. Previous approaches achieve step-to-step control using optimization by (1) using the exact model obtained by integrating the equations of motion, or (2) using a linear approximation of the step-to-step dynamics. The former provides a large region of stability at the expense of a high computational cost while the latter is computationally cheap but offers limited region of stability. Our method combines the advantages of both. First, we generate input/output data by simulating a single step. Second, the input/output data is curve fitted using a regression model to get a closed-form approximation of the step-to-step dynamics. We do this model identification offline. Next, we use the regression model for online optimal control. Here, using the spring-load inverted pendulum model of hopping, we show that both parametric (polynomial and neural network) and non-parametric (gaussian process regression) approximations can adequately model the step-to-step dynamics. We then show this approach can stabilize a wide range of initial conditions fast enough to enable real-time control. Our results suggest that closed-form approximation of the step-to-step dynamics provides a simple accurate model for fast optimal control of legged robots.
more »
« less
Optimal Control of a 5-Link Biped Using Quadratic Polynomial Model of Two-Point Boundary Value Problem.
To walk over constrained environments, bipedal robots must meet concise control objectives of speed and foot placement. The decisions made at the current step need to factor in their effects over a time horizon. Such step-to-step control is formulated as a two-point boundary value problem (2-BVP). As the dimensionality of the biped increases, it becomes increasingly difficult to solve this 2-BVP in real-time. The common method to use a simple linearized model for real-time planning followed by mapping on the high dimensional model cannot capture the nonlinearities and leads to potentially poor performance for fast walking speeds. In this paper, we present a framework for real-time control based on using partial feedback linearization (PFL) for model reduction, followed by a data-driven approach to find a quadratic polynomial model for the 2-BVP. This simple step-tostep model along with constraints is then used to formulate and solve a quadratically constrained quadratic program to generate real-time control commands. We demonstrate the efficacy of the approach in simulation on a 5-link biped following a reference velocity profile and on a terrain with ditches.
more »
« less
- Award ID(s):
- 1946282
- PAR ID:
- 10347376
- Date Published:
- Journal Name:
- In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
- Volume:
- 85451
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract This work pertains to the laser powder bed fusion (LPBF) additive manufacturing process. The goal of this work is to mitigate the expense and time required for qualification of laser powder bed fusion processed parts. In pursuit of this goal, the objective of this work is to develop and apply a physics-based model predictive control strategy to modulate the thermal history before the part is built. The key idea is to determine a desired thermal history for a given part a priori to printing using a physics-based model. Subsequently, a model predictive control strategy is developed to attain the desired thermal history by changing the laser power layer-by-layer. This is an important area of research because the spatiotemporal distribution of temperature within the part (also known as the thermal history) influences flaw formation, microstructure evolution, and surface/geometric integrity, all of which ultimately determine the mechanical properties of the part. Currently, laser powder bed fusion parts are qualified using a build-and-test approach wherein parameters are optimized by printing simple test coupons, followed by examining their properties via materials characterization and testing — a cumbersome and expensive process that often takes years. These parameters, once optimized, are maintained constant throughout the process for a part. However, thermal history is a function of over 50 processing parameters including material properties and part design, consequently, the current approach of parameter optimization based on empirical testing of simple test coupons seldom transfers successfully to complex, practical parts. Rather than instinctive process parameter optimization, the model predictive control strategy presents a radically different approach to LPBF part qualification that is based on understanding and modulating the causal thermal physics of the process. The approach has three steps: (Step 1) Predict – given a part geometry, use a rapid, mesh-less physics-based simulation model to predict its thermal history, analyze the predicted thermal history trend, isolate potential red flag problems such as heat buildup, and set a desired thermal history that corrects deleterious trends. (Step 2) Parse – iteratively simulate the thermal history as a function of various laser power levels layer-by-layer over a fixed time horizon. (Step 3) Select – the laser power that provides the closest match to the desired thermal history. Repeat Steps 2 and 3 until the part is completely built. We demonstrate through experiments with various geometries two advantages of this model predictive control strategy when applied to laser powder bed fusion: (i) prevent part failures due to overheating and distortion, while mitigating the need for anchoring supports; and (ii) improve surface integrity of hard to access internal surfaces.more » « less
-
null (Ed.)This paper presents a data-driven algorithm to solve the problem of infinite-horizon linear quadratic regulation (LQR), for a class of discrete-time linear time-invariant systems subjected to state and control constraints. The problem is divided into a constrained finite-horizon LQR subproblem and an unconstrained infinite-horizon LQR subproblem, which can be solved directly from collected input/state data, separately. Under certain conditions, the combination of the solutions of the subproblems converges to the optimal solution of the original problem. The effectiveness of the proposed approach is validated by a numerical example.more » « less
-
A control system for simulated two-dimensional bipedal walking was developed. The biped model was built based on anthropometric data. At the core of the control is a Deep Deterministic Policy Gradients (DDPG) neural network that is trained in GAZEBO, a physics simulator, to predict the ideal foot location to maintain stable walking under external impulse load. Additional controllers for hip joint movement during stance phase, and ankle joint torque during toeoff, help to stabilize the robot during walking. The simulated robot can walk at a steady pace of approximately 1m/s, and during locomotion it can maintain stability with a 30N-s impulse applied at the torso. This work implement DDPG algorithm to solve biped walking control problem. The complexity of DDPG network is decreased through carefully selected state variables and distributed control system.more » « less
-
Nonlinear optimal control problems are challenging to solve efficiently due to non-convexity. This paper introduces a trajectory optimization approach that achieves real-time performance by combining machine learning to predict optimal trajectories with refinement by quadratic optimization. First, a library of optimal trajectories is calculated offline and used to train a neural network. Online, the neural network predicts a trajectory for a novel initial state and cost function, and this prediction is further optimized by a sparse quadratic programming solver. We apply this approach to a fly-to-target movement problem for an indoor quadrotor. Experiments demonstrate that the technique calculates near-optimal trajectories in a few milliseconds, and generates agile movement that can be tracked more accurately than existing methods.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1115/DETC2021-70733
