Precision cutting of soft-tissue remains a challenging problem in robotics, due to the complex and unpredictable mechanical behaviour of tissue under manipulation. Here, we consider the challenge of cutting along the boundary between two soft mediums, a problem that is made extremely difficult due to visibility constraints, which means that the precise location of the cutting trajectory is typically unknown. This paper introduces a novel strategy to address this task, using a binary medium classifier trained using joint torque measurements, and a closed loop control law that relies on an error signal compactly encoded in the decision boundary of the classifier. We illustrate this on a grapefruit cutting task, successfully modulating a nominal trajectory fit using dynamic movement primitives to follow the boundary between grapefruit pulp and peel using torque based medium classification. Results show that this control strategy is successful in 72 % of attempts in contrast to control using a nominal trajectory, which only succeeds in 50 % of attempts.
Behaviour cloning is a commonly used strategy for imitation learning and can be extremely effective in constrained domains. However, in cases where the dynamics of an environment may be state dependent and varying, behaviour cloning places a burden on model capacity and the number of demonstrations required. This paper introduces switching density networks, which rely on a categorical reparametrisation for hybrid system identification. This results in a network comprising a classification layer that is followed by a regression layer. We use switching density networks to predict the parameters of hybrid control laws, which are toggled by a switching layer to produce different controller outputs, when conditioned on an input state. This work shows how switching density networks can be used for hybrid system identification in a variety of tasks, successfully identifying the key joint angle goals that make up manipulation tasks, while simultaneously learning image-based goal classifiers and regression networks that predict joint angles from images. We also show that they can cluster the phase space of an inverted pendulum, identifying the balance, spin and pump controllers required to solve this task. Switching density networks can be difficult to train, but we introduce a cross entropy regularisation loss that stabilises training.
Learning from demonstration is an effective method for human users to instruct desired robot behaviour. However, for most non-trivial tasks of practical interest, efficient learning from demonstration depends crucially on inductive bias in the chosen structure for rewards/costs and policies. We address the case where this inductive bias comes from an exchange with a human user. We propose a method in which a learning agent utilizes the information bottleneck layer of a high-parameter variational neural model, with auxiliary loss terms, in order to ground abstract concepts such as spatial relations. The concepts are referred to in natural language instructions and are manifested in the high-dimensional sensory input stream the agent receives from the world. We evaluate the properties of the latent space of the learned model in a photorealistic synthetic environment and particularly focus on examining its usability for downstream tasks. Additionally, through a series of controlled table-top manipulation experiments, we demonstrate that the learned manifold can be used to ground demonstrations as symbolic plans, which can then be executed on a PR2 robot.
Past research has studied two approaches to utilise predefined policy sets in repeated interactions: as experts, to dictate our own actions, and as types, to characterise the behaviour of other agents. In this work, we bring these complementary views together in the form of a novel meta-algorithm, called Expert-HBA (E-HBA), which can be applied to any expert algorithm that considers the average (or total) payoff an expert has yielded in the past. E-HBA gradually mixes the past payoff with a predicted future payoff, which is computed using the type-based characterisation. We present results from a comprehensive set of repeated matrix games, comparing the performance of several well-known expert algorithms with and without the aid of E-HBA. Our results show that E-HBA has the potential to significantly improve the performance of expert algorithms.
This paper is concerned with evaluating different multiagent learning (MAL) algorithms in problems where individual agents may be heterogenous, in the sense of utilizing different learning strategies, without the opportunity for prior agreements or information regarding coordination. Such a situation arises in ad hoc team problems, a model of many practical multiagent systems applications. Prior work in multiagent learning has often been focussed on homogeneous groups of agents, meaning that all agents were identical and a priori aware of this fact. Also, those algorithms that are specifically designed for ad hoc team problems are typically evaluated in teams of agents with fixed behaviours, as opposed to agents which are adapting their behaviours. In this work, we empirically evaluate five MAL algorithms, representing major approaches to multiagent learning but originally developed with the homogeneous setting in mind, to understand their behaviour in a set of ad hoc team problems. All teams consist of agents which are continuously adapting their behaviours. The algorithms are evaluated with respect to a comprehensive characterisation of repeated matrix games, using performance criteria that include considerations such as attainment of equilibrium, social welfare and fairness. Our main conclusion is that there is no clear winner. However, the comparative evaluation also highlights the relative strengths of different algorithms with respect to the type of performance criteria, e.g., social welfare vs. attainment of equilibrium.
Temporally extended and sequenced robot motion tasks are often characterized by discontinuous switches between different types of local dynamics. These change-points can be exploited to build approximate models of the interleaving regions, which in turn allow the design of region-specific controllers. These can then be combined to create the initiation state-space of a final policy. However, such a pipeline can become challenging to implement for combinatorially complex, temporarily extended tasks - especially so when sub-controllers work on different information streams, time scales and action spaces. In this paper, we introduce a method that can compose diverse policies based on scripted motion planning, dynamic motion primitives and neural networks. In order to do this, we extend the options framework to introduce a per-option dynamics module and a global function that evaluates a goal metric. Additionally, we can leverage expert demonstrations to sequence these local policies, converting the learning problem in hierarchical reinforcement learning to a planning problem at inference time. We first illustrate the core concepts with an MDP benchmark, and then with a physical gear assembly task solved on a PR2 robot. We show that the proposed approach successfully discovers the optimal sequence of policies and solves both tasks efficiently.
Robots performing tasks in dynamic environments would benefit greatly from understanding the underlying environment motion, in order to make future predictions and to synthesize effective control policies that use this inductive bias. Online system identification is therefore a fundamental requirement for robust autonomous agents. When the dynamics involves multiple modes (due to contacts or interactions between objects), and when system identification must proceed directly from a rich sensory stream such as video, then traditional methods for system identification may not be well suited. We propose an approach wherein fast parameter estimation with a model can be seamlessly combined with a recurrent variational autoencoder. Our Physics-based recurrent variational autoencoder model includes an additional loss that enforces conformity with the structure of a physically based dynamics model. This enables the resulting model to encode parameters such as position, velocity, restitution, air drag and other physical properties of the system. The model can be trained entirely in simulation, in an end-to-end manner with domain randomization, to perform online system identification, and probabilistic forward predictions of parameters of interest. We benchmark against existing system identification methods and demonstrate that Vid2Param outperforms the baselines in terms of speed and accuracy of identification, and also provides uncertainty quantification in the form of a distribution over future trajectories. Furthermore, we illustrate the utility of this in physical experiments wherein a PR2 robot with velocity constrained arm must intercept a bouncing ball, by estimating the physical parameters of this ball directly from the video trace after the ball is released.
While many multiagent algorithms are designed for homogeneous systems (i.e. all agents are identical), there are important applications which require an agent to coordinate its actions without knowing a priori how the other agents behave. One method to make this problem feasible is to assume that the other agents draw their latent policy (or type) from a specific set, and that a domain expert could provide a specification of this set, albeit only a partially correct one. Algorithms have been proposed by several researchers to compute posterior beliefs over such policy libraries, which can then be used to determine optimal actions. In this paper, we provide theoretical guidance on two central design parameters of this method: Firstly, it is important that the user choose a posterior which can learn the true distribution of latent types, as otherwise suboptimal actions may be chosen. We analyse convergence properties of two existing posterior formulations and propose a new posterior which can learn correlated distributions. Secondly, since the types are provided by an expert, they may be inaccurate in the sense that they do not predict the agents' observed actions. We provide a novel characterisation of optimality which allows experts to use efficient model checking algorithms to verify optimality of types.
Many multiagent applications require an agent to learn quickly how to interact with previously unknown other agents. To address this problem, researchers have studied learning algorithms which compute posterior beliefs over a hypothesised set of policies, based on the observed actions of the other agents. The posterior belief is complemented by the prior belief, which specifies the subjective likelihood of policies before any actions are observed. In this paper, we present the first comprehensive empirical study on the practical impact of prior beliefs over policies in repeated interactions. We show that prior beliefs can have a significant impact on the long-term performance of such methods, and that the magnitude of the impact depends on the depth of the planning horizon. Moreover, our results demonstrate that automatic methods can be used to compute prior beliefs with consistent performance effects. This indicates that prior beliefs could be eliminated as a manual parameter and instead be computed automatically.