While there has been substantial success in applying actor-critic methods to continuous control, simpler critic-only methods such as Q-learning often remain intractable in the associated high-dimensional action spaces. However, most actor-critic methods come at the cost of added complexity: heuristics for stabilization, compute requirements as well as wider hyperparameter search spaces. We show that these issues can be largely alleviated via Q-learning by combining action discretization with value decomposition, framing single-agent control as cooperative multi-agent reinforcement learning (MARL). With bang-bang actions, performance of this critic-only approach matches state-of-the-art continuous actor-critic methods when learning from features or pixels. We extend classical bandit examples from cooperative MARL to provide intuition for how decoupled critics leverage state information to coordinate joint optimization, and demonstrate surprisingly strong performance across a wide variety of continuous control tasks.
Experience replay plays a crucial role in improving the sample efficiency of deep reinforcement learning agents. Recent advances in experience replay propose using Mixup (Zhang et al., 2018) to further improve sample efficiency via synthetic sample generation. We build upon this technique with Neighborhood Mixup Experience Replay (NMER), a geometrically-grounded replay buffer that interpolates transitions with their closest neighbors in state-action space. NMER preserves a locally linear approximation of the transition manifold by only applying Mixup between transitions with vicinal state-action features. Under NMER, a given transition's set of state action neighbors is dynamic and episode agnostic, in turn encouraging greater policy generalizability via inter-episode interpolation. We combine our approach with recent off-policy deep reinforcement learning algorithms and evaluate on continuous control environments. We observe that NMER improves sample efficiency by an average 94% (TD3) and 29% (SAC) over baseline replay buffers, enabling agents to effectively recombine previous experiences and learn from limited data.
In most classical Autonomous Vehicle (AV) stacks, the prediction and planning layers are separated, limiting the planner to react to predictions that are not informed by the planned trajectory of the AV. This work presents a module that tightly couples these layers via a game-theoretic Model Predictive Controller (MPC) that uses a novel interactive multi-agent neural network policy as part of its predictive model. In our setting, the MPC planner considers all the surrounding agents by informing the multi-agent policy with the planned state sequence. Fundamental to the success of our method is the design of a novel multi-agent policy network that can steer a vehicle given the state of the surrounding agents and the map information. The policy network is trained implicitly with ground-truth observation data using backpropagation through time and a differentiable dynamics model to roll out the trajectory forward in time. Finally, we show that our multi-agent policy network learns to drive while interacting with the environment, and, when combined with the game-theoretic MPC planner, can successfully generate interactive behaviors.
Data-driven simulators promise high data-efficiency for driving policy learning. When used for modelling interactions, this data-efficiency becomes a bottleneck: Small underlying datasets often lack interesting and challenging edge cases for learning interactive driving. We address this challenge by proposing a simulation method that uses in-painted ado vehicles for learning robust driving policies. Thus, our approach can be used to learn policies that involve multi-agent interactions and allows for training via state-of-the-art policy learning methods. We evaluate the approach for learning standard interaction scenarios in driving. In extensive experiments, our work demonstrates that the resulting policies can be directly transferred to a full-scale autonomous vehicle without making use of any traditional sim-to-real transfer techniques such as domain randomization.
Simulation has the potential to transform the development of robust algorithms for mobile agents deployed in safety-critical scenarios. However, the poor photorealism and lack of diverse sensor modalities of existing simulation engines remain key hurdles towards realizing this potential. Here, we present VISTA, an open source, data-driven simulator that integrates multiple types of sensors for autonomous vehicles. Using high fidelity, real-world datasets, VISTA represents and simulates RGB cameras, 3D LiDAR, and event-based cameras, enabling the rapid generation of novel viewpoints in simulation and thereby enriching the data available for policy learning with corner cases that are difficult to capture in the physical world. Using VISTA, we demonstrate the ability to train and test perception-to-control policies across each of the sensor types and showcase the power of this approach via deployment on a full scale autonomous vehicle. The policies learned in VISTA exhibit sim-to-real transfer without modification and greater robustness than those trained exclusively on real-world data.
Reinforcement learning (RL) for continuous control typically employs distributions whose support covers the entire action space. In this work, we investigate the colloquially known phenomenon that trained agents often prefer actions at the boundaries of that space. We draw theoretical connections to the emergence of bang-bang behavior in optimal control, and provide extensive empirical evaluation across a variety of recent RL algorithms. We replace the normal Gaussian by a Bernoulli distribution that solely considers the extremes along each action dimension - a bang-bang controller. Surprisingly, this achieves state-of-the-art performance on several continuous control benchmarks - in contrast to robotic hardware, where energy and maintenance cost affect controller choices. Since exploration, learning,and the final solution are entangled in RL, we provide additional imitation learning experiments to reduce the impact of exploration on our analysis. Finally, we show that our observations generalize to environments that aim to model real-world challenges and evaluate factors to mitigate the emergence of bang-bang solutions. Our findings emphasize challenges for benchmarking continuous control algorithms, particularly in light of potential real-world applications.
Learning competitive behaviors in multi-agent settings such as racing requires long-term reasoning about potential adversarial interactions. This paper presents Deep Latent Competition (DLC), a novel reinforcement learning algorithm that learns competitive visual control policies through self-play in imagination. The DLC agent imagines multi-agent interaction sequences in the compact latent space of a learned world model that combines a joint transition function with opponent viewpoint prediction. Imagined self-play reduces costly sample generation in the real world, while the latent representation enables planning to scale gracefully with observation dimensionality. We demonstrate the effectiveness of our algorithm in learning competitive behaviors on a novel multi-agent racing benchmark that requires planning from image observations. Code and videos available at https://sites.google.com/view/deep-latent-competition.
Learning complex behaviors through interaction requires coordinated long-term planning. Random exploration and novelty search lack task-centric guidance and waste effort on non-informative interactions. Instead, decision making should target samples with the potential to optimize performance far into the future, while only reducing uncertainty where conducive to this objective. This paper presents latent optimistic value exploration (LOVE), a strategy that enables deep exploration through optimism in the face of uncertain long-term rewards. We combine finite horizon rollouts from a latent model with value function estimates to predict infinite horizon returns and recover associated uncertainty through ensembling. Policy training then proceeds on an upper confidence bound (UCB) objective to identify and select the interactions most promising to improve long-term performance. We apply LOVE to visual control tasks in continuous state-action spaces and demonstrate improved sample complexity on a selection of benchmarking tasks.
Deterministic neural networks (NNs) are increasingly being deployed in safety critical domains, where calibrated, robust and efficient measures of uncertainty are crucial. While it is possible to train regression networks to output the parameters of a probability distribution by maximizing a Gaussian likelihood function, the resulting model remains oblivious to the underlying confidence of its predictions. In this paper, we propose a novel method for training deterministic NNs to not only estimate the desired target but also the associated evidence in support of that target. We accomplish this by placing evidential priors over our original Gaussian likelihood function and training our NN to infer the hyperparameters of our evidential distribution. We impose priors during training such that the model is penalized when its predicted evidence is not aligned with the correct output. Thus the model estimates not only the probabilistic mean and variance of our target but also the underlying uncertainty associated with each of those parameters. We observe that our evidential regression method learns well-calibrated measures of uncertainty on various benchmarks, scales to complex computer vision tasks, and is robust to adversarial input perturbations.
Information gathering while interacting with other agents is critical in many emerging domains, such as self-driving cars, service robots, drone racing, and active surveillance. In these interactions, the interests of agents may be at odds with others, resulting in a non-cooperative dynamic game. Since unveiling one's own strategy to adversaries is undesirable, each agent must independently predict the other agents' future actions without communication. In the face of uncertainty from sensor and actuator noise, agents have to gain information over their own state, the states of others, and the environment. They must also consider how their own actions reveal information to others. We formulate this non-cooperative multi-agent planning problem as a stochastic dynamic game. Our solution uses local iterative dynamic programming in the belief space to find a Nash equilibrium of the game. We present three applications: active surveillance, guiding eyes for a blind agent, and autonomous racing. Agents with game-theoretic belief space planning win 44% more races compared to a baseline without game theory and 34% more than without belief space planning.