CoCo: Online Mixed-Integer Control Via Supervised Learning

Cauligi, Abhishek; Culbertson, Preston; Szczerbicki, Edward; Schwager, Mac; Stellato, Bartolomeo; Pavone, Marco

doi:10.1109/lra.2021.3135931

Cited by 20 publications

(18 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This will require a training dataset of high optimality and well-performed learning. Several learning schemes have been explored [10], [19].…”

Section: A Data-driven Methods For Minlpsmentioning

confidence: 99%

“…Where w v , w θ , w h are the weights for velocity tracking, rotation angle tracking and body height tracking. For walking forward, we first set w v = [100, 100, 10], w θ = [10,10,10], w h = 10. This set of weight favors more on the forward walking speed.…”

Section: A Forward Walkingmentioning

confidence: 99%

“…Consequently, the problem formulation incorporates non-convex multi-linear terms for dynamics and binary terms for gait selection, hence a mixedinteger non-convex optimization problem (MINLP). In the previous paper [8], we have compared several equivalent approaches to resolve this problem including the nonlinear alternating direction method of multipliers (ADMM) approach [9], the data-driven approach to warm-start mixed-integer programs or mathematical programs with complementary constraints online, and the data-driven approach to solve the convex programmings online [10]. The convex programming approach shows potential to solve the problem with several trials of convex programming given well-performed supervised learning.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Learning Near-global-optimal Strategies for Hybrid Non-convex Model Predictive Control of Single Rigid Body Locomotion

Li¹,

Xing²,

Schperberg³

et al. 2022

Preprint

View full text Add to dashboard Cite

Convex model predictive controls (MPCs) with a single rigid body model have demonstrated strong performance on real legged robots. However, convex MPCs are limited by their assumptions such as small rotation angle and pre-defined gait, limiting the richness of potential solutions. We remove those assumptions and solve the complete mixed-integer nonconvex programming with single rigid body model. We first collect datasets of pre-solved problems offline, then learn the problem-solution map to solve this optimization fast for MPC. If warm-starts can be found, offline problems can be solved close to the global optimality. The proposed controller is tested by generating various gaits and behaviors depending on the initial conditions. Hardware test demonstrates online gait generation and adaptation running at more than 50 Hz based on sensor feedback.

show abstract

“…This will require a training dataset of high optimality and well-performed learning. Several learning schemes have been explored [10], [19].…”

Section: A Data-driven Methods For Minlpsmentioning

confidence: 99%

Section: A Forward Walkingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Near-global-optimal Strategies for Hybrid Non-convex Model Predictive Control of Single Rigid Body Locomotion

Li¹,

Xing²,

Schperberg³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…For architecture search, Luo et al (2018) learn a continuous latent space behind the discrete architecture space. Many reinforcement learning and control methods over discrete spaces can also be seen as amortizing or semi-amortizing the discrete control problems, for example: Cauligi et al (2020Cauligi et al ( , 2021…”

Section: Amortized Optimization Over Discrete Domainsmentioning

confidence: 99%

Tutorial on amortized optimization for learning to optimize over continuous domains

Amos¹

2022

Preprint

View full text Add to dashboard Cite

Optimization is a ubiquitous modeling tool and is often deployed in settings which repeatedly solve similar instances of the same problem. Amortized optimization methods use learning to predict the solutions to problems in these settings. This leverages the shared structure between similar problem instances. In this tutorial, we will discuss the key design choices behind amortized optimization, roughly categorizing 1) models into fully-amortized and semi-amortized approaches, and 2) learning methods into regression-based and objectivebased. We then view existing applications through these foundations to draw connections between them, including for manifold optimization, variational inference, sparse coding, meta-learning, control, reinforcement learning, convex optimization, and deep equilibrium networks. This framing enables us easily see, for example, that the amortized inference in variational autoencoders is conceptually identical to value gradients in control and reinforcement learning as they both use fully-amortized models with an objective-based loss.

show abstract

“…In the context of machine learning, CoCo proposed in [14] finds feasible solution to MIP by first learning to map the problem parameters to the assignment of the discrete variables offline and then solving the resulted continuous optimization problem online. While this greatly improves the solution speed at test time, it assumes that one is able to solve the original MIP in a reasonable amount of time to construct the dataset.…”

Section: Introductionmentioning

confidence: 99%

Efficient Object Manipulation Planning with Monte Carlo Tree Search

Zhu¹,

Righetti²

2022

Preprint

View full text Add to dashboard Cite

This paper presents an efficient approach to object manipulation planning using Monte Carlo Tree Search (MCTS) to find contact sequences and an efficient ADMM-based trajectory optimization algorithm to evaluate the dynamic feasibility of candidate contact sequences. To accelerate MCTS, we propose a methodology to learn a goal-conditioned policy-value network used to direct the search towards promising nodes. Further, manipulation-specific heuristics enable to drastically reduce the search space. Systematic object manipulation experiments in a physics simulator demonstrate the efficiency of our approach. In particular, our approach scales favorably for long manipulation sequences thanks to the learned policy-value network, significantly improving planning success rate.

show abstract

CoCo: Online Mixed-Integer Control Via Supervised Learning

Cited by 20 publications

References 14 publications

Learning Near-global-optimal Strategies for Hybrid Non-convex Model Predictive Control of Single Rigid Body Locomotion

Learning Near-global-optimal Strategies for Hybrid Non-convex Model Predictive Control of Single Rigid Body Locomotion

Tutorial on amortized optimization for learning to optimize over continuous domains

Efficient Object Manipulation Planning with Monte Carlo Tree Search

Contact Info

Product

Resources

About