Abstract— Robotic prostheses can improve walking ability in persons with transfemoral amputations by closely matching kinetics and kinematics of the intact leg during walking. However, achieving this goal requires the prosthesis to adapt to walking speed, a function that no powered transfemoral prosthesis has yet achieved. In this paper, we present, and perform initial testing on a new control framework that allows biologically accurate leg function at varying walking speeds, without the need for tuning. The proposed framework comprises two novel controllers that rely on quasi-stiffness modulation in stance phase and minimum jerk trajectory in swing phase. Preliminary testing was conducted in an able- bodied subject using a bypass adapter to walk on a robotic prosthesis at five different walking speeds (from 0.62 to 1.16 m/s). Experimental results demonstrated the ability of the proposed controller to approximate intact leg function at different walking speeds. I. INTRODUCTION Every year, 185,000 people undergo a major lower-limb amputation in the United States [1]. Functional impairment following loss of a leg is particularly severe for persons with transfemoral amputations, who expend up to twice the metabolic effort to walk at half the speed of able-bodied persons [2], and experience a higher risk of falls and incidence of secondary conditions such as back pain, depression, and osteoarthritis [3]. Experimental studies and gait simulations show that passive prostheses cannot fully replace the biomechanical function of intact legs, as they are unable to provide biologically-accurate torque at the joint level [4]. Persons with amputations must compensate for this deficiency by increased effort of both the residual limb and contralateral leg [5]. However, these compensatory strategies result in kinetic and kinematics gait asymmetries that reduce gait efficiency and cause higher stress on the musculoskeletal apparatus [5]. Thanks to recent improvements on motor, battery, and microcontroller technologies, robotic prostheses have become a reality [6-9]. Robotic prostheses can actively regulate joint torque and release positive net-energy during the gait cycle. Thus they can closely match the kinetics and kinematics of an intact leg during walking, possibly restoring physiological gait efficiency and stability. Research supported by USAMRAA grant # W81XWH-09-2-0020. T. Lenzi and L. J. Hargrove is with the Center for Bionic Medicine at the Rehabilitation Institute of Chicago and the Department of Physical Medicine and Rehabilitation of Northwestern University, Chicago, IL, USA (phone: 312-238-1570; e-mail: lenzi@ieee.org). J. W. Sensinger was with the Rehabilitation Institute of Chicago, Chicago, IL, 60611, USA. He is now with the Institute of Biomedical Engineering, University of New Brunswick, Canada (e-mail: sensinger@ieee.org). Despite promising results obtained so far [10-11], several challenges remain in the design and control of a robotic prosthesis. From a control perspective, one of the main challenges is to provide biologically accurate torque across a wide range of walking speeds, without requiring speed-specific prosthesis tuning, which is unfeasible in clinical practice [12]. Joint torque demand is not fixed but depends on walking speed [13]. In stance phase (i.e., foot on the ground), torque profiles change non-linearly with walking speed to properly support body weight against gravity, and to propel the body mass in a forward direction [14]. In swing phase (i.e., foot off the ground), a progressively faster movement must be generated at increased walking speeds to ensure timely placement of the foot in preparation for the subsequent heel-strike, regardless of the initial swing conditions generated by the stance controller. Available transfemoral prosthesis controllers cannot adapt to different walking speeds. Impedance-inspired control [15] defines the joint torque as a parametric function of joint angle and velocity, with different stiffness, damping, and equilibrium values for each discrete phase of the gait cycle. To obtain biologically-accurate torques, impedance- inspired control requires explicit tuning of all parameters through experiments performed at a specific walking speed [15]. At speeds other than that specific speed, prosthesis function is not optimal; it provides incorrect body support and propulsion during stance phase and incorrect movement trajectory and duration in swing phase. Most importantly, because of speed-dependent parameters, impedance-inspired control requires user-specific tuning. Different persons use different gait cadences—and therefore different joint velocities—to walk at the same speed. A non-linear control approach has been recently proposed to address the speed adaptability issue in stance phase [16], though actual tests at varying walking speeds have not yet been presented [17]. Effective speed adaptation has been achieved in ankle- foot prostheses using two different control approaches. Herr [10] proposed a biophysically inspired function that, by imitating muscle reflexes, allows the device to obtain inherent speed-adaptability (i.e., joint torque is adapted without a direct measure of walking speed or cadence). The method proposed by Holgate et al. [18] relied instead on pre- programmed ankle torque profiles based on a continuous estimate of the gait cycle progression and an explicit measure of gait pace. Although an extension of these control approaches to transfemoral prostheses would be possible in stance phase, they still do not address the need for speed adaptability in swing phase, which is fundamental for transfemoral prostheses. In this paper, we propose a new control framework that allows robotic transfemoral prostheses to generate Preliminary Evaluation of a New Control Approach to Achieve Speed Adaptation in Robotic Transfemoral Prostheses T. Lenzi, Member, IEEE, L.J. Hargrove, Member, IEEE, and J. W. Sensinger, Member, IEEE