Max Robotics — Global Service Robot Directory + US Certification

About

Alexander Toshev is a prominent robotics and AI researcher whose work sits at the intersection of embodied intelligence, robot navigation, and language-grounded control. His research focuses on enabling robots to understand and act upon high-level human instructions, navigate complex real-world environments, and perform sophisticated manipulation tasks through learned visual and linguistic representations. Toshev's most celebrated contribution is his co-authorship of "Do As I Can, Not As I Say" (2022), which demonstrated how large language models can be grounded in real-world robotic affordances to enable nuanced, instruction-following behavior — a landmark paper accumulating over 516 citations that has significantly shaped the direction of language-conditioned robotics. His work on Scene Memory Transformers (186 citations) advanced how embodied agents handle long-horizon tasks in partially observable settings, while his research on sim-to-real visual servoing (108 citations) tackled the challenge of viewpoint-invariant robotic control using recurrent neural networks. Toshev has also made meaningful contributions to social robot navigation, helping establish evaluation standards and large-scale datasets like SCAND that the broader community relies upon. Across his portfolio, spanning reinforcement learning, semantic navigation, and human-robot interaction, his research has collectively garnered over 1,200 citations, reflecting sustained and far-reaching impact on modern robotics.