arXiv subjects

Nelson Lojo

Publications and source records attributed to Nelson Lojo.

3 recordsLinked to original sources

arXiv2025

Control of Microrobots with Reinforcement Learning under On-Device Compute Constraints

An important function of autonomous microrobots is the ability to perform robust movement over terrain. This paper explores an edge ML approach to microrobot locomotion, allowing for on-device, lower latency control under compute, memory, and power constraints. This paper explores the locomotion of a sub-centimeter quadrupedal microrobot via reinforcement learning (RL) and deploys the resulting controller on an ultra-small system-on-chip (SoC), SC$\mu$M-3C, featuring an ARM Cortex-M0 microcontroller running at 5 MHz. We train a compact FP32 multilayer perceptron (MLP) policy with two hidden layers ($[128, 64]$) in a massively parallel GPU simulation and enhance robustness by utilizing domain randomization over simulation parameters. We then study integer (Int8) quantization (per-tensor and per-feature) to allow for higher inference update rates on our resource-limited hardware, and we connect hardware power budgets to achievable update frequency via a cycles-per-update model for inference on our Cortex-M0. We propose a resource-aware gait scheduling viewpoint: given a device power budget, we can select the gait mode (trot/intermediate/gallop) that maximizes expected RL reward at a corresponding feasible update frequency. Finally, we deploy our MLP policy on a real-world large-scale robot on uneven terrain, qualitatively noting that domain-randomized training can improve out-of-distribution stability. We do not claim real-world large-robot empirical zero-shot transfer in this work.

cs.RO↗

arXiv2025

Using Large Language Models to Develop Requirements Elicitation Skills

Requirements Elicitation (RE) is a crucial software engineering skill that involves interviewing a client and then devising a software design based on the interview results. Teaching this inherently experiential skill effectively has high cost, such as acquiring an industry partner to interview, or training course staff or other students to play the role of a client. As a result, a typical instructional approach is to provide students with transcripts of real or fictitious interviews to analyze, which exercises the skill of extracting technical requirements but fails to develop the equally important interview skill itself. As an alternative, we propose conditioning a large language model to play the role of the client during a chat-based interview. We perform a between-subjects study (n=120) in which students construct a high-level application design from either an interactive LLM-backed interview session or an existing interview transcript describing the same business processes. We evaluate our approach using both a qualitative survey and quantitative observations about participants' work. We find that both approaches provide sufficient information for participants to construct technically sound solutions and require comparable time on task, but the LLM-based approach is preferred by most participants. Importantly, we observe that LLM-backed interview is seen as both more realistic and more engaging, despite the LLM occasionally providing imprecise or contradictory information. These results, combined with the wide accessibility of LLMs, suggest a new way to practice critical RE skills in a scalable and realistic manner without the overhead of arranging live interviews.

cs.SE↗

arXiv2024

Can Custom Models Learn In-Context? An Exploration of Hybrid Architecture Performance on In-Context Learning Tasks

In-Context Learning (ICL) is a phenomenon where task learning occurs through a prompt sequence without the necessity of parameter updates. ICL in Multi-Headed Attention (MHA) with absolute positional embedding has been the focus of more study than other sequence model varieties. We examine implications of architectural differences between GPT-2 and LLaMa as well as LlaMa and Mamba. We extend work done by Garg et al. (2022) and Park et al. (2024) to GPT-2/LLaMa hybrid and LLaMa/Mamba hybrid models - examining the interplay between sequence transformation blocks and regressive performance in-context. We note that certain architectural changes cause degraded training efficiency/ICL accuracy by converging to suboptimal predictors or converging slower. We also find certain hybrids showing optimistic performance improvements, informing potential future ICL-focused architecture modifications. Additionally, we propose the "ICL regression score", a scalar metric describing a model's whole performance on a specific task. Compute limitations impose restrictions on our architecture-space, training duration, number of training runs, function class complexity, and benchmark complexity. To foster reproducible and extensible research, we provide a typed, modular, and extensible Python package on which we run all experiments.

cs.LG↗