A5005267157- Reinforcement Learning in Robotics
- Advanced Bandit Algorithms Research
- Data Stream Mining Techniques
- Adversarial Robustness in Machine Learning
- Optimization and Search Problems
- Robotic Path Planning Algorithms
- Advanced Control Systems Optimization
- Evolutionary Algorithms and Applications
- Robot Manipulation and Learning
- Video Surveillance and Tracking Methods
- Teaching and Learning Programming
- Tensor decomposition and applications
- Markov Chains and Monte Carlo Methods
- Advanced Vision and Imaging
Carnegie Mellon University
2018-2020
Case Western Reserve University
2016-2017
We present a method for fast training of vision based control policies on real robots.The key idea behind our is to perform multi-task Reinforcement Learning with auxiliary tasks that differ not only in the reward be optimized but also state-space which they operate.In particular, we allow task utilize features are available at training-time.This allows learning policies, subsequently generate good data main, vision-based policies.This can seen as an extension Scheduled Auxiliary Control...
Achieving effective task performance on real mobile robots is a great challenge when hand-coding algorithms, both due to the amount of effort involved and manually tuned parameters required for each skill. Learning algorithms instead have potential lighten up this by using one single set training learning different skills, but question feasibility such in remains research pursuit. We focus kind robot system - soccer "small-size" domain, which tactical high-level team strategies build upon...
We explore using reinforcement learning on single and multi-agent systems such that after is finished we can apply a policy zero-shot to new environment sizes, as well different number of agents entities. Building off previous work, show how map back forth between the state action space standard Markov Decision Process (MDP) multi-dimensional tensors transfer in these cases possible. Like use special network architecture designed work with tensor representation, known Fully Convolutional...
We present a method for fast training of vision based control policies on real robots. The key idea behind our is to perform multi-task Reinforcement Learning with auxiliary tasks that differ not only in the reward be optimized but also state-space which they operate. In particular, we allow task utilize features are available at training-time. This allows learning policies, subsequently generate good data main, vision-based policies. can seen as an extension Scheduled Auxiliary Control...