Robot Air Hockey: A Manipulation Testbed for Robot Learning with Reinforcement Learning

Reinforcement Learning is a promising tool for learning complex policies even in fast-moving and object-interactive domains where human teleoperation or hard-coded policies might fail. To effectively reflect this challenging category of tasks, we introduce a dynamic, interactive RL testbed based on robot air hockey. By augmenting air hockey with a large family of tasks ranging from easy tasks like reaching, to challenging ones like pushing a block by hitting it with a puck, as well as goal-based and human-interactive tasks, our testbed allows a varied assessment of RL capabilities. The robot air hockey testbed also supports sim-to-real transfer with three domains: two simulators of increasing fidelity and a real robot system. Using a dataset of demonstration data gathered through two teleoperation systems: a virtualized control environment, and human shadowing, we assess the testbed with behavior cloning, offline RL, and RL from scratch.

Using Machine Learning to Reduce Observational Biases When Detecting New Impacts on Mars

The current inventory of recent (fresh) impacts on Mars shows a strong bias towards areas of low thermal inertia. These areas are generally visually bright, and impacts create dark scours and rays that make them easier to detect. It is expected that impacts occur at a similar rate in areas of higher thermal inertia, but those impacts are under-detected. This study investigates the use of a trained machine learning classifier to increase the detection of fresh impacts on Mars using CTX data. This approach discovered 69 new fresh impacts that have been confirmed with follow-up HiRISE images. We found that examining candidates partitioned by thermal inertia (TI) values, which is only possible due to the large number of machine learning candidates, helps reduce the observational bias and increase the number of known high-TI impacts.