filmov
tv
Multi-agent Reinforcement Learning in Urban and Search Rescue: RI Summer Scholar Long Le

Показать описание
Multi-agent Hierarchical Reinforcement Learning in Urban and Search Rescue
Longe Le, RISS 2021 Cohort
In an urban search-and-rescue (USAR) task, a team of agents cooperates to explore different rooms in the environment, clear rubble, and triage victims. USAR presents a challenging control problem in multi-agent Reinforcement Learning due to its long horizon, sparse and delayed reward, and the large size of the state space. Hierarchical approaches such as options and state abstraction have been proposed as a way to reduce the state space and planning complexity in such problems. In this work, we leverage domain knowledge to (1) train individual low-level (sub)-policies of each agent in smaller subsets of the state space, and (2) then train a team-level policy over a reduced graph representation of the states using those pretrained sub-policies. We show some early-stage results where our approach outperforms two multi-agent algorithms option-critic, and independent Q-learning on a simple environment.
QUICK DATES AND INFORMATION:
- Applications are open November 1
- Program runs June 1 through the first week of August
- Scholarships for both U.S. and International Undergraduate Students Available
- Hosted by the Robotics Institute at Carnegie Mellon University, Pittsburgh
SCHOLARSHIPS: The RISS program partners and community provides extensive scholarships for students selected to participate. Scholarships are available to students from across the US and world.
RESEARCH RESULTS:
- Scholars contribute to robotics research projects through a guided research experience with multiple layers of mentorship
- Scholars learn how to effectively communicate research ideas to various audiences (e.g., sponsors, academic audience, novice audiences) and in various formats (e.g., elevator pitches, short talks, research papers, and poster presentations).
- Scholars forge long-lasting connections to Carnegie Mellon University researchers and partners.
Robotics Workshops & Talks: The technical professional development series exposes scholars to a wide range of robotics applications and projects through weekly robotics talks, visits to labs, and hands- on workshops.
Longe Le, RISS 2021 Cohort
In an urban search-and-rescue (USAR) task, a team of agents cooperates to explore different rooms in the environment, clear rubble, and triage victims. USAR presents a challenging control problem in multi-agent Reinforcement Learning due to its long horizon, sparse and delayed reward, and the large size of the state space. Hierarchical approaches such as options and state abstraction have been proposed as a way to reduce the state space and planning complexity in such problems. In this work, we leverage domain knowledge to (1) train individual low-level (sub)-policies of each agent in smaller subsets of the state space, and (2) then train a team-level policy over a reduced graph representation of the states using those pretrained sub-policies. We show some early-stage results where our approach outperforms two multi-agent algorithms option-critic, and independent Q-learning on a simple environment.
QUICK DATES AND INFORMATION:
- Applications are open November 1
- Program runs June 1 through the first week of August
- Scholarships for both U.S. and International Undergraduate Students Available
- Hosted by the Robotics Institute at Carnegie Mellon University, Pittsburgh
SCHOLARSHIPS: The RISS program partners and community provides extensive scholarships for students selected to participate. Scholarships are available to students from across the US and world.
RESEARCH RESULTS:
- Scholars contribute to robotics research projects through a guided research experience with multiple layers of mentorship
- Scholars learn how to effectively communicate research ideas to various audiences (e.g., sponsors, academic audience, novice audiences) and in various formats (e.g., elevator pitches, short talks, research papers, and poster presentations).
- Scholars forge long-lasting connections to Carnegie Mellon University researchers and partners.
Robotics Workshops & Talks: The technical professional development series exposes scholars to a wide range of robotics applications and projects through weekly robotics talks, visits to labs, and hands- on workshops.