Real-world robotics tasks commonly manifest ascontrol problems overcontinuous action spaces. When learning to act in such settings, control policies are typically represented as continuous probability distributions that cover all feasible control inputs - often Gaussians. The underlying assumption is that this enables more refined decisions compared to crude policy choices such as discretized controllers, which limit the search space but induce abrupt changes. While switching controls canbeundesirable inpractice astheymaychallenge stability andaccelerate system weardown, they are theoretically feasible and even arise as optimal strategies in some settings.

Theexpected performance objective in (13) is equivalent setting toβ = 0 in Figure 1a. The algorithm gets stuck and repeatedly evaluates inputs (orange crosses) at alocal optimum of the true objectivefunction (black dashed).

The action cost differs by row, where the corresponding optimal policy parameterization is highlighted in green.

We consider the simplest possible case that still allows for nonlinear dynamics.

We thank the reviewer for pointing this out.

The action cost differs by row, where the corresponding optimal policy parameterization is highlighted in green.

We present a novel reinforcement learning (RL) environment designed to both optimize industrial sorting systems and study agent behavior in evolving spaces. In simulating material flow within a sorting process our environment follows the idea of a digital twin, with operational parameters like belt speed and occupancy level. To reflect real-world challenges, we integrate common upgrades to industrial setups, like new sensors or advanced machinery. It thus includes two variants: a basic version focusing on discrete belt speed adjustments and an advanced version introducing multiple sorting modes and enhanced material composition observations. We detail the observation spaces, state update mechanisms, and reward functions for both environments. We further evaluate the efficiency of common RL algorithms like Proximal Policy Optimization (PPO), Deep-Q-Networks (DQN), and Advantage Actor Critic (A2C) in comparison to a classical rule-based agent (RBA). This framework not only aids in optimizing industrial processes but also provides a foundation for studying agent behavior and transferability in evolving environments, offering insights into model performance and practical implications for real-world RL applications.

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.