We explore the viability of casting foundation models as generic reward functions for reinforcement learning. To this end, we propose a simple pipeline that interfaces an off-the-shelf vision model with a large language model. Specifically, given a trajectory of observations, we infer the likelihood of an instruction describing the task that the user wants an agent to perform. We show that this generic likelihood function exhibits the characteristics ideally expected from a reward function: it associates high values with the desired behaviour and lower values for several similar, but incorrect policies. Overall, our work opens the possibility of designing open-ended agents for interactive tasks via foundation models.

Training autonomous agents that can learn new tasks from only a handful of demonstrations is a long-standing problem in machine learning. Recently, transformers have been shown to learn new language or vision tasks without any weight updates from only a few examples, also referred to as in-context learning. However, the sequential decision making setting poses additional challenges having a lower tolerance for errors since the environment's stochasticity or the agent's actions can lead to unseen, and sometimes unrecoverable, states. In this paper, we use an illustrative example to show that naively applying transformers to sequential decision making problems does not enable in-context learning of new tasks. We then demonstrate how training on sequences of trajectories with certain distributional properties leads to in-context learning of new sequential decision making tasks. We investigate different design choices and find that larger model and dataset sizes, as well as more task diversity, environment stochasticity, and trajectory burstiness, all result in better in-context learning of new out-of-distribution tasks. By training on large diverse offline datasets, our model is able to learn new MiniHack and Procgen tasks without any weight updates from just a handful of demonstrations.

We study the choice of action space in robot manipulation learning and sim-to-real transfer. We define metrics that assess the performance, and examine the emerging properties in the different action spaces. We train over 250 reinforcement learning~(RL) agents in simulated reaching and pushing tasks, using 13 different control spaces. The choice of action spaces spans popular choices in the literature as well as novel combinations of common design characteristics. We evaluate the training performance in simulation and the transfer to a real-world environment. We identify good and bad characteristics of robotic action spaces and make recommendations for future designs. Our findings have important implications for the design of RL algorithms for robot manipulation tasks, and highlight the need for careful consideration of action spaces when training and transferring RL agents for real-world robotics.

In emergency scenarios, mobile robots must navigate like humans, interpreting stimuli to locate potential victims rapidly without interfering with first responders. Existing socially-aware navigation algorithms face computational and adaptability challenges. To overcome these, we propose a solution, MIRACLE -- an inverse reinforcement and curriculum learning model, that employs gamified learning to gather stimuli-driven human navigational data. This data is then used to train a Deep Inverse Maximum Entropy Reinforcement Learning model, reducing reliance on demonstrator abilities. Testing reveals a low loss of 2.7717 within a 400-sized environment, signifying human-like response replication. Current databases lack comprehensive stimuli-driven data, necessitating our approach. By doing so, we enable robots to navigate emergency situations with human-like perception, enhancing their life-saving capabilities.

While recent years have seen rapid progress in image-conditioned text generation, image captioning still suffers from the fundamental issue of hallucinations, the generation of spurious details that cannot be inferred from the given image. Dedicated methods for reducing hallucinations in image captioning largely focus on closed-vocabulary object tokens, ignoring most types of hallucinations that occur in practice. In this work, we propose MOCHa, an approach that harnesses advancements in reinforcement learning (RL) to address the sequence-level nature of hallucinations in an open-world setup. To optimize for caption fidelity to the input image, we leverage ground-truth reference captions as proxies to measure the logical consistency of generated captions. However, optimizing for caption fidelity alone fails to preserve the semantic adequacy of generations; therefore, we propose a multi-objective reward function that jointly targets these qualities, without requiring any strong supervision. We demonstrate that these goals can be simultaneously optimized with our framework, enhancing performance for various captioning models of different scales. Our qualitative and quantitative results demonstrate MOCHa's superior performance across various established metrics. We also demonstrate the benefit of our method in the open-vocabulary setting. To this end, we contribute OpenCHAIR, a new benchmark for quantifying open-vocabulary hallucinations in image captioning models, constructed using generative foundation models. We will release our code, benchmark, and trained models.

In GCD, the joint training of EBM and a diffusion model is formulated as a minimax problem, which reaches an equilibrium when both models converge to the data distribution. The minimax learning with GCD bears interesting equivalence to inverse reinforcement learning, where the energy corresponds to a negative reward, the diffusion model is a policy, and the real data is expert demonstrations. We present preliminary yet promising results showing that joint training is beneficial for both EBM and a diffusion model. GCD enables EBM training without MCMC while improving the sample quality of a diffusion model.

Addressing decision-making problems using sequence modeling to predict future trajectories shows promising results in recent years. In this paper, we take a step further to leverage the sequence predictive method in wider areas such as long-term planning, vision-based control, and multi-task decision-making. To this end, we propose a method to utilize a diffusion-based generative sequence model to plan a series of milestones in a latent space and to have an agent to follow the milestones to accomplish a given task. The proposed method can learn control-relevant, low-dimensional latent representations of milestones, which makes it possible to efficiently perform long-term planning and vision-based control. Furthermore, our approach exploits generation flexibility of the diffusion model, which makes it possible to plan diverse trajectories for multi-task decision-making. We demonstrate the proposed method across offline reinforcement learning (RL) benchmarks and an visual manipulation environment. The results show that our approach outperforms offline RL methods in solving long-horizon, sparse-reward tasks and multi-task problems, while also achieving the state-of-the-art performance on the most challenging vision-based manipulation benchmark.

The reinforcement learning algorithms have often been applied to social robots. However, most reinforcement learning algorithms were not optimized for the use of social robots, and consequently they may bore users. We proposed a new reinforcement learning method specialized for the social robot, the FRAC-Q-learning, that can avoid user boredom. The proposed algorithm consists of a forgetting process in addition to randomizing and categorizing processes. This study evaluated interest and boredom hardness scores of the FRAC-Q-learning by a comparison with the traditional Q-learning. The FRAC-Q-learning showed significantly higher trend of interest score, and indicated significantly harder to bore users compared to the traditional Q-learning. Therefore, the FRAC-Q-learning can contribute to develop a social robot that will not bore users. The proposed algorithm can also find applications in Web-based communication and educational systems. This paper presents the entire process, detailed implementation and a detailed evaluation method of the of the FRAC-Q-learning for the first time.

Multi-agent planning and reinforcement learning can be challenging when agents cannot see the state of the world or communicate with each other due to communication costs, latency, or noise. Partially Observable Stochastic Games (POSGs) provide a mathematical framework for modelling such scenarios. This paper aims to improve the efficiency of planning and reinforcement learning algorithms for POSGs by identifying the underlying structure of optimal state-value functions. The approach involves reformulating the original game from the perspective of a trusted third party who plans on behalf of the agents simultaneously. From this viewpoint, the original POSGs can be viewed as Markov games where states are occupancy states, i.e., posterior probability distributions over the hidden states of the world and the stream of actions and observations that agents have experienced so far. This study mainly proves that the optimal state-value function is a convex function of occupancy states expressed on an appropriate basis in all zero-sum, common-payoff, and Stackelberg POSGs.

We take the first step in studying general sequential decision-making under two adaptivity constraints: rare policy switch and batch learning. First, we provide a general class called the Eluder Condition class, which includes a wide range of reinforcement learning classes. Then, for the rare policy switch constraint, we provide a generic algorithm to achieve a $\widetilde{\mathcal{O}}(\log K) $ switching cost with a $\widetilde{\mathcal{O}}(\sqrt{K})$ regret on the EC class. For the batch learning constraint, we provide an algorithm that provides a $\widetilde{\mathcal{O}}(\sqrt{K}+K/B)$ regret with the number of batches $B.$ This paper is the first work considering rare policy switch and batch learning under general function classes, which covers nearly all the models studied in the previous works such as tabular MDP (Bai et al. 2019; Zhang et al. 2020), linear MDP (Wang et al. 2021; Gao et al. 2021), low eluder dimension MDP (Kong et al. 2021; Gao et al. 2021), generalized linear function approximation (Qiao et al. 2023), and also some new classes such as the low $D_\Delta$-type Bellman eluder dimension problem, linear mixture MDP, kernelized nonlinear regulator and undercomplete partially observed Markov decision process (POMDP).