We investigated the cavity length and temperature dependence of the device characteristics for 1.5 µm InGaAsP/InGaAsP multiple quantum well lasers with a two-step compositional separate confinement heterostructure by eptiaxial regrowth. The device parameters, such as internal optical loss, internal quantum efficiency, transparency current density and modal differential gain, were estimated by fittings to experimental values of threshold current density and slope efficiency as a function of cavity length. Under continuous wave (CW) mode at 25 • C, the uncoated 600 µm long laser emitted a maximum output power of 22 mW with a threshold current of 9.3 mA and a slope efficiency of 0.17 mW mA −1 , and it operated up to 97 • C. The full-width at half-maximum values of far-field patterns with a single lobe remained almost constant, indicating ∼18.7 • (horizontal) × 24 • (vertical) at 25 • C over a wide range of injection current. In terms of maximum CW operating temperature, the optimum cavity length is governed by the trade off between emitting volume and internal temperature of lasers, leading to a temperature of 116 • C at a cavity length of 900 µm. By applying a high-reflectivity coating, based on an Au/Ti metallic mirror with an insulation layer of SiO 2 , on the rear facet of lasers, the device performance was significantly improved due to the reduced mirror loss without degradation of electrical properties, resulting in a high output power of 36 mW and a low threshold current of 7.8 mA and with a slope efficiency of 0.263 mW mA −1 under CW mode at 25 • C for the 600 µm long cavity.

Reinforcement learning agents that have not been seen during training must be robust in test environments. However, the generalization problem is challenging to solve in reinforcement learning using high-dimensional images as the input. The addition of a self-supervised learning framework with data augmentation in the reinforcement learning architecture can promote generalization to a certain extent. However, excessively large changes in the input images may disturb reinforcement learning. Therefore, we propose a contrastive learning method that can help manage the trade-off relationship between the performance of reinforcement learning and auxiliary tasks against the data augmentation strength. In this framework, strong augmentation does not disturb reinforcement learning and instead maximizes the auxiliary effect for generalization. Results of experiments on the DeepMind Control suite demonstrate that the proposed method effectively uses strong data augmentation and achieves a higher generalization than the existing methods.