This paper introduces a lightweight knee assisting robot called 'COWALK-M' and a control methodology using estimated ground slope and intended walking speed of the wearer. It is designed to assist stroke patients suffering from mild hemiplegia to move the paretic knee joint during Activities of Daily Living(ADL). The COWALK-M has various kinds of sensors including pressure sensors under the soles as well as encoders at the knee and ankle joints. These sensors are used to detect gait phases in real-time and finite state machine (FSM) is implemented for quiet standing and forward walking. Furthermore, ground slope as well as gait speed is estimated in order to provide the wearer with the most appropriate trajectory. To evaluate the performance of gait segmentation algorithm and gait speed/ground slope estimator, experiments on walking with and without a treadmill were conducted on five subjects under different speeds and different inclines. Despite a small number of sensors used, our method showed relatively high success rates for phase detection and a small error for estimators.