Multiagent learning offers a rich framework to address challenging real-world problems such as remote exploration and healthcare coordination, which require autonomous agents to express elaborate interactions. To be effective in such systems, agents must collectively reason about and pursue high-level, long-term, and possibly nebulous objectives while adapting their strategy to...
Reinforcement learning has made impressive strides in solving problems in challenging domains, but problems are increasingly being described with sparse rewards. Sparse rewards directly reduce the rate at which useful feedback is provided to the learner and make it difficult to distinguish between what specific actions led to the reception...
Casting manipulation in robotics is the act of launching a tethered projectile and using it to interact with the environment. One such interaction is wrapping a target. In this work, we identify throwing and tether control parameters for successfully wrapping a target with an end-weighted cable, evolving through a set...
There is growing commercial interest in the use of multiagent systems in real world applications. Some examples include inventory management in warehouses, smart homes, planetary exploration, search and rescue, air-traffic management and autonomous transportation systems. However, multiagent coordination is an extremely challenging problem. First, information relevant for coordination is often...
Legged locomotion in robotics benefits from engineered hardware dynamics such as configuration-dependent, nonlinear transmissions and compliance. This thesis introduces a novel cable driven leg that allows the design of configuration-dependent transmission ratios while also permitting the addition of compliance in line with the cables to achieve compliant legged locomotion. The...
We present a method for decentralized, multi-robot exploration in adverse environments where communication is minimal. A key conceptual feature of our method is enabling implicit coordination between robots by training a Convolutional Neural Network (CNN) as a heuristic for planning using Monte Carlo Tree Search (MCTS). Our method consists of...
In recent years, model-free Deep Reinforcement Learning (RL) has become an increasingly popular alternative to more traditional model-based or optimization-based control methods in solving robotic legged locomotion. However, deploying RL in the real world can be a significant undertaking. Constructing reward functions which compel controllers to learn the desired behavior...
Many animals and robots move through the world by coupling cyclical changes in shape called gaits to an interaction with the environment. Because mobility is an important aspect of such robots, a key metric when evaluating design and performance of mobile robots is the efficiency of their optimal gaits. The...
Multi-robot teams offer promising solutions for many long term deployments in remote and dangerous domains, such as extraterrestrial or underseas exploration. However, long term deployments present many problems preventing robot teams from operating effectively. Learning over long time scales is makes it difficult to assign credit to robots' actions, as...
As robots are becoming more relevant to our lives, they are still having hard time accomplishing simple tasks such as picking and lifting. Problems that include environmental constraints, pose uncertainties and hardware noises restrain robots for grasping an object successfully from a perceivable environment. Many have looked into finding best...