Offline data are both valuable and practical resources for teaching robots complex behaviors. Ideally, learning agents should not be constrained by the scarcity of available demonstrations, but rather generalize beyond the training distribution. However, the complexity of real-world scenarios typically requires huge amounts of data to prevent neural network policies from picking up on spurious correlations and learning non-causal relationships. We propose CAIAC, a data augmentation method that can create feasible synthetic transitions from a fixed dataset without having access to online environment interactions. By utilizing principled methods for quantifying causal influence, we are able to perform counterfactual reasoning by swapping $\textit{action}$-unaffected parts of the state-space between independent trajectories in the dataset. We empirically show that this leads to a substantial increase in robustness of offline learning algorithms against distributional shift.

Real-world robotic manipulation tasks remain an elusive challenge, since they involve both fine-grained environment interaction, as well as the ability to plan for long-horizon goals. Although deep reinforcement learning (RL) methods have shown encouraging results when planning end-to-end in high-dimensional environments, they remain fundamentally limited by poor sample efficiency due to inefficient exploration, and by the complexity of credit assignment over long horizons. In this work, we present Efficient Learning of High-Level Plans from Play (ELF-P), a framework for robotic learning that bridges motion planning and deep RL to achieve long-horizon complex manipulation tasks. We leverage task-agnostic play data to learn a discrete behavioral prior over object-centric primitives, modeling their feasibility given the current context. We then design a high-level goal-conditioned policy which (1) uses primitives as building blocks to scaffold complex long-horizon tasks and (2) leverages the behavioral prior to accelerate learning. We demonstrate that ELF-P has significantly better sample efficiency than relevant baselines over multiple realistic manipulation tasks and learns policies that can be easily transferred to physical hardware.

Although many real-time neural decoding algorithms have been proposed for brain-machine interface (BMI) applications over the years, an optimal, consensual approach remains elusive. Recent advances in deep learning algorithms provide new opportunities for improving the design of BMI decoders, including the use of recurrent artificial neural networks to decode neuronal ensemble activity in real time. Here, we developed a long-short term memory (LSTM) decoder for extracting movement kinematics from the activity of large ( N = 134–402) populations of neurons, sampled simultaneously from multiple cortical areas, in rhesus monkeys performing motor tasks. Recorded regions included primary motor, dorsal premotor, supplementary motor, and primary somatosensory cortical areas. The LSTM's capacity to retain information for extended periods of time enabled accurate decoding for tasks that required both movements and periods of immobility. Our LSTM algorithm significantly outperformed the state-of-the-art unscented Kalman filter when applied to three tasks: center-out arm reaching, bimanual reaching, and bipedal walking on a treadmill. Notably, LSTM units exhibited a variety of well-known physiological features of cortical neuronal activity, such as directional tuning and neuronal dynamics across task epochs. LSTM modeled several key physiological attributes of cortical circuits involved in motor tasks. These findings suggest that LSTM-based approaches could yield a better algorithm strategy for neuroprostheses that employ BMIs to restore movement in severely disabled patients.