Searcharxiv⌕ Search

arXiv subjects

Akhil S. Anand

Publications and source records attributed to Akhil S. Anand.

2 recordsLinked to original sources

Gradient-Based Co-Design of Systems and Reinforcement Learning Policies with Differentiable Simulators

Co-design is the problem of jointly optimizing the design of a system and the policy that operates it. Neither can be chosen independently: a system design is only as good as the policy trained for it, and the optimal policy changes with the design. When that policy is learned by reinforcement learning (RL), evaluating a candidate design requires a full policy training run, and the design is usually optimized by black-box search such as Bayesian optimization (BO) or evolution strategies (ES). These methods scale poorly with the design dimension and ignore the differentiable structure of modern simulators and, in particular, the sensitivity of the optimal policy with respect to the design. We instead derive a gradient-based co-design method, where the design gradient decomposes into a direct effect of changing the design at a fixed policy and an indirect effect through the design-dependent optimal policy. The indirect effect requires the sensitivity of the converged policy to the design, which we obtain by applying the implicit function theorem to the stationarity condition of the policy gradient. Based on this sensitivity, we develop a proof-of-concept co-design algorithm with stochastic value gradient (SVG)-style policy optimization, prove that it converges to an approximately stationary design, and demonstrate on a differentiable networked-system benchmark with high-dimensional design spaces that it consistently outperforms BO and ES.

cs.LG↗

Economic Model Predictive Control as a Solution to Markov Decision Processes

Markov Decision Processes (MDPs) offer a fairly generic and powerful framework to discuss the notion of optimal policies for dynamic systems, in particular when the dynamics are stochastic. However, computing the optimal policy of an MDP can be very difficult due to the curse of dimensionality present in solving the underlying Bellman equations. Model Predictive Control (MPC) is a very popular technique for building control policies for complex dynamic systems. Historically, MPC has focused on constraint satisfaction and steering dynamic systems towards a user-defined reference. More recently, Economic MPC was proposed as a computationally tractable way of building optimal policies for dynamic systems. When stochsaticity is present, economic MPC is close to the MDP framework. In that context, Economic MPC can be construed as attractable heuristic to provide approximate solutions to MDPs. However, there is arguably a knowledge gap in the literature regarding these approximate solutions and the conditions for an MPC scheme to achieve closed-loop optimality. This chapter aims to clarify this approximation pedagogically, to provide the conditions for MPC to deliver optimal policies, and to explore some of their consequences.

eess.SY↗