Real-world robotic tasks require complex reward functions. When we define the problem the robot needs to solve, we pretend that a designer specifies this complex reward exactly, and it is set in stone from then on. In practice, however, reward design is an iterative process: the designer chooses a reward, eventually encounters an "edge-case" environment where the reward incentivizes the wrong behavior, revises the reward, and repeats. What would it mean to rethink robotics problems to formally account for this iterative nature of reward design? We propose that the robot not take the specified reward for granted, but rather have uncertainty about it, and account for the future design iterations as future evidence. We contribute an Assisted Reward Design method that speeds up the design process by anticipating and influencing this future evidence: rather than letting the designer eventually encounter failure cases and revise the reward then, the method actively exposes the designer to such environments during the development phase. We test this method in a simplified autonomous driving task and find that it more quickly improves the car's behavior in held-out environments by proposing environments that are "edge cases" for the current reward.
相關內容
Logics with analogous semantics, such as Fuzzy Logic, have a number of explanatory and application advantages, the most well-known being the ability to help experts develop control systems. From a cognitive systems perspective, such languages also have the advantage of being grounded in perception. For social decision making in humans, it is vital that logical conclusions about others (cognitive empathy) are grounded in empathic emotion (affective empathy). Classical Fuzzy Logic, however, has several disadvantages: it is not obvious how complex formulae, e.g., the description of events in a text, can be (a) formed, (b) grounded, and (c) used in logical reasoning. The two-layered Context Logic (CL) was designed to address these issue. Formally based on a lattice semantics, like classical Fuzzy Logic, CL also features an analogous semantics for complex fomulae. With the Activation Bit Vector Machine (ABVM), it has a simple and classical logical reasoning mechanism with an inherent imagery process based on the Vector Symbolic Architecture (VSA) model of distributed neuronal processing. This paper adds to the existing theory how scales, as necessary for adjective and verb semantics can be handled by the system.
Legged robots that can operate autonomously in remote and hazardous environments will greatly increase opportunities for exploration into under-explored areas. Exteroceptive perception is crucial for fast and energy-efficient locomotion: perceiving the terrain before making contact with it enables planning and adaptation of the gait ahead of time to maintain speed and stability. However, utilizing exteroceptive perception robustly for locomotion has remained a grand challenge in robotics. Snow, vegetation, and water visually appear as obstacles on which the robot cannot step~-- or are missing altogether due to high reflectance. Additionally, depth perception can degrade due to difficult lighting, dust, fog, reflective or transparent surfaces, sensor occlusion, and more. For this reason, the most robust and general solutions to legged locomotion to date rely solely on proprioception. This severely limits locomotion speed, because the robot has to physically feel out the terrain before adapting its gait accordingly. Here we present a robust and general solution to integrating exteroceptive and proprioceptive perception for legged locomotion. We leverage an attention-based recurrent encoder that integrates proprioceptive and exteroceptive input. The encoder is trained end-to-end and learns to seamlessly combine the different perception modalities without resorting to heuristics. The result is a legged locomotion controller with high robustness and speed. The controller was tested in a variety of challenging natural and urban environments over multiple seasons and completed an hour-long hike in the Alps in the time recommended for human hikers.
This thesis is concerned with deriving planning algorithms for robot manipulators. Manipulation has two effects, the robot has a physical effect on the object, and it also acquires information about the object. This thesis presents algorithms that treat both problems. First, I present an extension of the well-known piano mover's problem where a robot pushing an object must plan its movements as well as those of the object. This requires simultaneous planning in the joint space of the robot and the configuration space of the object, in contrast to the original problem which only requires planning in the latter space. The effects of a robot action on the object configuration are determined by the non-invertible rigid body mechanics. Second, I consider planning under uncertainty and in particular planning for information effects. I consider the case where a robot has to reach and grasp an object under pose uncertainty caused by shape incompleteness. The approach presented in this report is to study and possibly extend a new approach to artificial intelligence (A.I.) which has emerged in the last years in response to the necessity of building intelligent controllers for agents operating in unstructured stochastic environments. Such agents require the ability to learn by interaction with its environment an optimal action-selection behaviour. The main issue is that real-world problems are usually dynamic and unpredictable. Thus, the agent needs to update constantly its current image of the world using its sensors, which provide only a noisy description of the surrounding environment. Although there are different schools of thinking, with their own set of techniques, a brand new direction which unifies many A.I. researches is to formalise such agent/environment interactions as embedded systems with stochastic dynamics.
Integrated sensing and communication (ISAC) has been regarded as one of the most promising technologies for future wireless communications. However, the mutual interference in the communication radar coexistence system cannot be ignored. Inspired by the studies of reconfigurable intelligent surface (RIS), we propose a double-RIS-assisted coexistence system where two RISs are deployed for enhancing communication signals and suppressing mutual interference. We aim to jointly optimize the beamforming of RISs and radar to maximize communication performance while maintaining radar detection performance. The investigated problem is challenging, and thus we transform it into an equivalent but more tractable form by introducing auxiliary variables. Then, we propose a penalty dual decomposition (PDD)-based algorithm to solve the resultant problem. Moreover, we consider two special cases: the large radar transmit power scenario and the low radar transmit power scenario. For the former, we prove that the beamforming design is only determined by the communication channel and the corresponding optimal joint beamforming strategy can be obtained in closed-form. For the latter, we minimize the mutual interference via the block coordinate descent (BCD) method. By combining the solutions of these two cases, a low-complexity algorithm is also developed. Finally, simulation results show that both the PDD-based and low-complexity algorithms outperform benchmark algorithms.
Active inference is a unifying theory for perception and action resting upon the idea that the brain maintains an internal model of the world by minimizing free energy. From a behavioral perspective, active inference agents can be seen as self-evidencing beings that act to fulfill their optimistic predictions, namely preferred outcomes or goals. In contrast, reinforcement learning requires human-designed rewards to accomplish any desired outcome. Although active inference could provide a more natural self-supervised objective for control, its applicability has been limited because of the shortcomings in scaling the approach to complex environments. In this work, we propose a contrastive objective for active inference that strongly reduces the computational burden in learning the agent's generative model and planning future actions. Our method performs notably better than likelihood-based active inference in image-based tasks, while also being computationally cheaper and easier to train. We compare to reinforcement learning agents that have access to human-designed reward functions, showing that our approach closely matches their performance. Finally, we also show that contrastive methods perform significantly better in the case of distractors in the environment and that our method is able to generalize goals to variations in the background.
For an autonomous agent to fulfill a wide range of user-specified goals at test time, it must be able to learn broadly applicable and general-purpose skill repertoires. Furthermore, to provide the requisite level of generality, these skills must handle raw sensory input such as images. In this paper, we propose an algorithm that acquires such general-purpose skills by combining unsupervised representation learning and reinforcement learning of goal-conditioned policies. Since the particular goals that might be required at test-time are not known in advance, the agent performs a self-supervised "practice" phase where it imagines goals and attempts to achieve them. We learn a visual representation with three distinct purposes: sampling goals for self-supervised practice, providing a structured transformation of raw sensory inputs, and computing a reward signal for goal reaching. We also propose a retroactive goal relabeling scheme to further improve the sample-efficiency of our method. Our off-policy algorithm is efficient enough to learn policies that operate on raw image observations and goals for a real-world robotic system, and substantially outperforms prior techniques.
This paper proposes a Reinforcement Learning (RL) algorithm to synthesize policies for a Markov Decision Process (MDP), such that a linear time property is satisfied. We convert the property into a Limit Deterministic Buchi Automaton (LDBA), then construct a product MDP between the automaton and the original MDP. A reward function is then assigned to the states of the product automaton, according to accepting conditions of the LDBA. With this reward function, our algorithm synthesizes a policy that satisfies the linear time property: as such, the policy synthesis procedure is "constrained" by the given specification. Additionally, we show that the RL procedure sets up an online value iteration method to calculate the maximum probability of satisfying the given property, at any given state of the MDP - a convergence proof for the procedure is provided. Finally, the performance of the algorithm is evaluated via a set of numerical examples. We observe an improvement of one order of magnitude in the number of iterations required for the synthesis compared to existing approaches.
The Normalized Cut (NCut) objective function, widely used in data clustering and image segmentation, quantifies the cost of graph partitioning in a way that biases clusters or segments that are balanced towards having lower values than unbalanced partitionings. However, this bias is so strong that it avoids any singleton partitions, even when vertices are very weakly connected to the rest of the graph. Motivated by the B\"uhler-Hein family of balanced cut costs, we propose the family of Compassionately Conservative Balanced (CCB) Cut costs, which are indexed by a parameter that can be used to strike a compromise between the desire to avoid too many singleton partitions and the notion that all partitions should be balanced. We show that CCB-Cut minimization can be relaxed into an orthogonally constrained $\ell_{\tau}$-minimization problem that coincides with the problem of computing Piecewise Flat Embeddings (PFE) for one particular index value, and we present an algorithm for solving the relaxed problem by iteratively minimizing a sequence of reweighted Rayleigh quotients (IRRQ). Using images from the BSDS500 database, we show that image segmentation based on CCB-Cut minimization provides better accuracy with respect to ground truth and greater variability in region size than NCut-based image segmentation.
A vexing problem in artificial intelligence is reasoning about events that occur in complex, changing visual stimuli such as in video analysis or game play. Inspired by a rich tradition of visual reasoning and memory in cognitive psychology and neuroscience, we developed an artificial, configurable visual question and answer dataset (COG) to parallel experiments in humans and animals. COG is much simpler than the general problem of video analysis, yet it addresses many of the problems relating to visual and logical reasoning and memory -- problems that remain challenging for modern deep learning architectures. We additionally propose a deep learning architecture that performs competitively on other diagnostic VQA datasets (i.e. CLEVR) as well as easy settings of the COG dataset. However, several settings of COG result in datasets that are progressively more challenging to learn. After training, the network can zero-shot generalize to many new tasks. Preliminary analyses of the network architectures trained on COG demonstrate that the network accomplishes the task in a manner interpretable to humans.
Deep reinforcement learning (RL) methods generally engage in exploratory behavior through noise injection in the action space. An alternative is to add noise directly to the agent's parameters, which can lead to more consistent exploration and a richer set of behaviors. Methods such as evolutionary strategies use parameter perturbations, but discard all temporal structure in the process and require significantly more samples. Combining parameter noise with traditional RL methods allows to combine the best of both worlds. We demonstrate that both off- and on-policy methods benefit from this approach through experimental comparison of DQN, DDPG, and TRPO on high-dimensional discrete action environments as well as continuous control tasks. Our results show that RL with parameter noise learns more efficiently than traditional RL with action space noise and evolutionary strategies individually.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191