Aleksandra Malysheva
Papers
1
Total Citations
3
H-Index
1
About
Aleksandra Malysheva is a researcher in artificial intelligence and robotics, with a focus on reinforcement learning and humanoid locomotion. Her work addresses the fundamental challenge of teaching simulated agents complex motor skills from scratch—a notoriously difficult problem due to high-dimensional action spaces and sparse rewards. In her highly regarded paper, "Learning to Run with Potential-Based Reward Shaping and Demonstrations from Video Data" (2018), Malysheva proposed an innovative approach that combines reward shaping with video-based demonstrations to accelerate learning. This work was directly inspired by the NIPS 2017 "Learning to Run" competition, where she tackled the task of training a two-legged humanoid model to navigate a simulated race course. By leveraging potential-based reward functions and extracting movement priors from video data, her method significantly improved sample efficiency and policy quality. While her citation count is still growing, her contributions are notable for bridging the gap between demonstration learning and reward engineering in continuous control. Malysheva’s research is particularly relevant for students and researchers interested in robot learning, sim-to-real transfer, and the intersection of computer vision and reinforcement learning for motor skill acquisition.
Research Focus
Key Achievements
Top Papers
- 1