Sharad Vikram

About

Sharad Vikram is a researcher at the forefront of model-based reinforcement learning (RL), with a particular focus on enabling data-efficient control in complex, high-dimensional environments. His most influential work, "SOLAR: Deep Structured Representations for Model-Based Reinforcement Learning" (2018), has garnered 131 citations and introduced a novel method for learning structured latent representations that make iterative model-based planning tractable from raw image observations. This contribution directly addresses a core challenge in RL: bridging the gap between sample efficiency and perceptual complexity. Vikram has also made important strides in the sim-to-real transfer problem, co-authoring "How to pick the domain randomization parameters for sim-to-real transfer of reinforcement learning policies?" (2019), which provides practical guidance for deploying RL policies trained in simulation onto physical systems. His work is characterized by a rigorous, principled approach to representation learning, aiming to build RL systems that are both theoretically sound and practically deployable. By tackling the dual challenges of representation quality and real-world transfer, Vikram’s research is helping to lay the groundwork for more robust and generalizable autonomous agents.

Research Focus

Key Achievements

Top Papers

1. 1
   SOLAR: Deep Structured Representations for Model-Based Reinforcement Learning

   131 citations · 2018
2. 2
   How to pick the domain randomization parameters for sim-to-real transfer of reinforcement learning policies?

   22 citations · 2019

Key Collaborators

Contact & Links