Efficient Language-Guided Reinforcement Learning for Resource-Constrained Autonomous Systems

Abstract

In this paper, we propose an energy-efficient architecture which is designed to receive both images and text inputs as a step towards designing reinforcement learning agents that can understand human language and act in real-world environments. We evaluate our proposed method on three different software environments and a low power drone named Crazyflie to navigate towards specified goals and avoid obstacles successfully. To find the most efficient language-guided RL model, we implemented the model with various configurations of image input sizes and text instruction sizes on the Crazyflie drone GAP8 which consists of 8 RISC-V cores. The task completion success rate and onboard power consumption, latency, and memory usage of GAP8 are measured and compared with Jetson TX2 ARM CPU and Raspberry Pi 4. The results show that by decreasing 20% of input image size we achieve up to 78% energy improvement while achieving an 82% task completion success rate.

MLAE2: Metareasoning for Latency-Aware Energy-Efficient Autonomous Nano-Drones

“…We used two pre-trained models in [5] as tiny machine learning and cloud models. Figure 4 (b) and (c) depict two different setups for the power measurement including a drone [5], [6] used in this paper and GAPuino [24], [25], [26] which has the GAP8 processor [27]. After extracting offline power consumption and latency, we used them in the simulation while implementing the proposed algorithm in this paper.…”

“…Tiny drones have equipped with various sensors such as LIDAR and cameras and the collected data by these sensors can be fed to ML Neural Networks (NNs). Vision-based Deep Neural Networks (DNNs) [3], [5], [4] or Reinforcement Learning (RL) [6], [7], [8], [9], [10] approaches can be deployed on such tiny drones to enable them to perform complex tasks. Due to the intensive computational requirements of DNNs models, cloud-based approaches which provide un-limited computational capacity have been highly addressed in this area.…”

“…Fig. 4: (a) Crazyflie with AI-Deck, (b) drone power measurement setup[5],[6], and (c) power measurement setup includes GAPuino with camera[24].…”